WO2022205618A1 - 定位方法、行驶控制方法、装置、计算机设备及存储介质 - Google Patents

定位方法、行驶控制方法、装置、计算机设备及存储介质 Download PDF

Info

Publication number
WO2022205618A1
WO2022205618A1 PCT/CN2021/098964 CN2021098964W WO2022205618A1 WO 2022205618 A1 WO2022205618 A1 WO 2022205618A1 CN 2021098964 W CN2021098964 W CN 2021098964W WO 2022205618 A1 WO2022205618 A1 WO 2022205618A1
Authority
WO
WIPO (PCT)
Prior art keywords
target object
target
image
geometric feature
dimensional geometric
Prior art date
Application number
PCT/CN2021/098964
Other languages
English (en)
French (fr)
Inventor
唐庆
王潇峰
刘余钱
Original Assignee
上海商汤临港智能科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 上海商汤临港智能科技有限公司 filed Critical 上海商汤临港智能科技有限公司
Publication of WO2022205618A1 publication Critical patent/WO2022205618A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30248—Vehicle exterior or interior
    • G06T2207/30252—Vehicle exterior; Vicinity of vehicle
    • G06T2207/30256—Lane; Road marking

Definitions

  • the present disclosure relates to the technical field of computer vision, and in particular, to a positioning method, a driving control method, an apparatus, a computer device, and a storage medium.
  • determining the vehicle's pose information is a very basic link.
  • tasks such as vehicle perception, decision-making, and path planning need to be further processed on the basis of determining the current vehicle's pose information. Therefore, The accuracy of the determined pose information directly affects the accuracy of subsequent perception, the rationality of decision-making, and the fineness of planning.
  • the embodiments of the present disclosure provide at least a positioning method, a driving control method, an apparatus, a computer device, and a storage medium.
  • an embodiment of the present disclosure provides a positioning method, including: acquiring a target image obtained by collecting a target scene; and determining, based on the target image, that a first target object included in the target image is in the target The first two-dimensional geometric feature in the image; based on the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene, the first two-dimensional geometric feature of the first target object , and the corresponding relationship between the first target object and the second target object, perform position matching on the first target object and the second target object; Collect the target pose information of the device.
  • the matching process is simple, and the amount of data processed is less, so the processing speed is faster, and the determination can be improved.
  • the efficiency of the target image acquisition device's target pose information is simple, and the amount of data processed is less, so the processing speed is faster, and the determination can be improved.
  • the first two-dimensional geometric feature of the first target object when the first target object includes a first road sign object with a rectangular outline, the first two-dimensional geometric feature of the first target object includes: the first the vertex of the target object; in the case where the first target object includes a second road sign object with an irregular contour, the first two-dimensional geometric feature of the first target object includes: the contour of the first target object Lines and/or corners; in the case that the first target object includes a line-type third road sign object, the first two-dimensional geometric features of the first target object include: belonging to the first target object The target line segment located in the image area and located on the center line of the first target object.
  • the first two-dimensional geometric features of different first target objects can be easily identified and more accurate. represents the actual position of the first target object in the target scene.
  • the determining, based on the target image, the first two-dimensional geometric feature of the first target object included in the target image in the target image includes: Perform semantic segmentation processing to determine the semantic segmentation results corresponding to the plurality of pixels in the target image respectively; based on the semantic segmentation results corresponding to the plurality of pixels respectively, and the respective positions of the plurality of pixels in the target image , and determine the first two-dimensional geometric feature of the first target object in the target image.
  • the semantic segmentation results corresponding to multiple pixels in the target image can be relatively accurately determined when the target image is subjected to semantic segmentation processing, the semantic segmentation results corresponding to multiple pixels are used to determine multiple When the position of the pixel point in the target image is obtained, the obtained first two-dimensional geometric feature is more accurate.
  • the semantic segmentation results corresponding to the plurality of pixel points and the The positions of the points in the target image respectively, and determining the first two-dimensional geometric feature of the first target object in the target image, including: based on the semantic segmentation result, determining from the target image that the the pixel points of the outline of the first target object; based on the pixel points belonging to the outline of the first target object, the corresponding bounding box of the first target object in the target image is obtained by fitting; based on the bounding box The vertices of determine the first two-dimensional geometric features of the first target object in the target image.
  • the semantic segmentation results corresponding to the plurality of pixel points, and the The respective positions of the pixel points in the target image, and determining the first two-dimensional geometric feature of the first target object in the target image includes: based on the semantic segmentation result, determining from the target image the pixel points of the outline of the first target object; based on the position of the pixel points belonging to the outline of the first target object in the target image, the outline of the first target object is obtained; based on the first target object The contour line of the target object determines the first two-dimensional geometric feature of the first target object in the target image.
  • the contour line of the first target object can be determined, which reduces the difficulty of vertex identification caused by the irregular edge of the contour line determined based on the pixel points,
  • the expression of the first two-dimensional geometric feature of the first target object is simplified to a large extent, which is beneficial to the matching between the same target objects.
  • the first two-dimensional geometric feature of the first target object is obtained through the determined target line segment of the first target object, which can better solve the problem that the stop line and the solid lane line continue to appear on the road, only the The problem of determining the contour of the stop line and the solid line of the lane cannot accurately calculate the position and attitude of the autonomous vehicle, so as to more accurately determine the first two-dimensional geometric feature of the first target object.
  • the method further includes: based on the three-dimensional geometric feature of the second target object in the target scene in the target scene and the first two-dimensional geometric feature of the first target object, generating the the corresponding relationship between the first target object and the second target object.
  • the three-dimensional geometric features of the second target object in the target scene and the first two-dimensional geometric features of the first target object are relatively simple geometric features, the three-dimensional geometric features and the first two-dimensional geometric features are used When the geometric feature generates the corresponding relationship between the first target object and the second target object, the amount of calculation is small, and the processing speed is fast, so the processing efficiency can be improved.
  • the generation of the The corresponding relationship between the first target object and the second target object includes: based on the initial pose information of the capture device that captures the target image, and the second target object in the target scene. three-dimensional geometric features, project the second target object into the image coordinate system of the target image, and obtain the first projected geometric feature of the second target object in the image coordinate system; based on the first target The first two-dimensional geometric feature of the object in the image coordinate system, and the first projected geometric feature of the second target object in the image coordinate system, for the first target object and the second target The objects are matched to obtain the corresponding relationship between the first target object and the second target object.
  • the determined first projected geometric feature and the first two-dimensional geometric feature are matched.
  • the matching since the first projected geometric feature and the corresponding geometric features of the first two-dimensional geometric feature are relatively simple, the matching
  • the corresponding relationship between the first target object and the second target object can be determined easily by matching in the same image coordinate system.
  • the generation of the The corresponding relationship between the first target object and the second target object includes: based on the homography matrix between the target image and the target plane where the second target object is located, the first target image in the target image. Projecting the target object into the target plane to obtain the second projected geometric feature of the first target object in the target plane; based on the second projected geometric feature of the first target object in the target plane, and the geometric features of the second target object in the target plane, the first target object and the second target object are matched to obtain the correspondence between the first target object and the second target object relationship; wherein, the geometric feature of the second target object in the target plane is determined based on the three-dimensional geometric feature of the second target object in the scene coordinate system.
  • the homography matrix can be used to project the first target object to the transformation matrix in the target plane more accurately
  • the second projected geometric feature obtained by projecting the first target object to the target plane is also more accurate. This makes it possible to use the second projected geometric feature of the first target object in the target plane and the geometric feature of the second target object in the target plane to match the first target object and the second target object more accurately. to obtain the correspondence between the first target object and the second target object.
  • the second object is characterized by the features representing the vertices of the first target object in the second projected geometric features and the geometric features of the second target object in the target plane
  • the features of the vertices determined by the object are matched to obtain the corresponding relationship between the first target object and the second target object; in the case that the first target object includes a second road sign object with an irregular outline, use
  • the features representing the contour lines and/or corners of the first target object in the second projected geometric features, and the geometric features of the second target object in the target plane representing the second target object The characteristics of the determined contour lines and/or corner points are matched to obtain the corresponding relationship between the first target object and the second target object; in
  • the determining, based on the result of the position matching, the target pose information of the collecting device that collects the target image includes: determining a position matching error based on the result of the position matching; The error and the initial pose information of the capture device that captures the target image determine the target pose information of the capture device that captures the target image.
  • the position matching error can be used to reflect the accuracy of the pose information of the target image acquisition device, and the pose information is continuously optimized based on the matching loss to improve the accuracy of the target pose information.
  • the determining the target pose information of the acquisition device that collects the target image based on the position matching error and the initial pose information of the acquisition device that collects the target image includes: detecting: Whether the preset iteration stop condition is met; if the iteration stop condition is met, the initial pose information obtained in the last iteration is determined as the target pose information; if the iteration stop condition is not met In the case of , based on the position matching error and the initial pose information in the latest iteration process, determine new initial pose information, and return to the target scene based on the second target object in the target scene.
  • the corresponding three-dimensional geometric features in the scene coordinate system, the first two-dimensional geometric features of the first target object, and the corresponding relationship between the first target object and the second target object, for the first target object The step of performing position matching with the second target object.
  • the obtained target pose information can achieve a higher confidence level, that is, the obtained target pose information is more accurate.
  • the iteration stop condition includes any one of the following: the number of iterations is greater than a preset number of iterations threshold; the position matching error between the first target object and the second target object is less than a preset value. loss threshold.
  • the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene, the first two-dimensional geometry of the first target object feature, and the corresponding relationship between the first target object and the second target object, and performing position matching on the first target object and the second target object includes: In the case of a second road sign object with a regular outline, interpolation processing is performed on the first two-dimensional geometric feature of the first target object to obtain the second two-dimensional geometric feature of the first target object; wherein the first two-dimensional geometric feature of the first target object is obtained;
  • the two-dimensional geometric features include: multiple vertices and multiple interpolation points; based on the second two-dimensional geometric features, the three-dimensional geometric features, and the correspondence between the first target object and the second target object, Perform point-to-point position matching on the first target object and the second target object.
  • the weights between different semantics can be balanced, and at the same time, the matching inconsistency can be alleviated when using fewer vertices when performing position matching processing. good question.
  • the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene, the first two-dimensional geometry of the first target object feature, and the corresponding relationship between the first target object and the second target object, and performing position matching on the first target object and the second target object includes: based on the acquisition of the target image The initial pose information of the device and the three-dimensional geometric features of the second target object in the target scene in the scene coordinate system corresponding to the target scene, and project the second target object to the image coordinates of the target image In the image coordinate system, the third projected geometric feature of the second target object in the image coordinate system is obtained; based on the third projected geometric feature of the second target object in the image coordinate system, the third projected geometric feature A first two-dimensional geometric feature of a target object is used to perform position matching on the first target object and the second target object having a corresponding relationship.
  • an embodiment of the present disclosure provides a driving control method for an intelligent driving device, including: acquiring video frame data collected by the intelligent driving device during driving; using the first aspect or any optional method of the first aspect
  • the positioning method in the embodiment processes the video frame data, detects a target object in the video frame data, and controls the intelligent driving device based on the detected target object.
  • the positioning method provided by the embodiment of the present disclosure can be used to determine the target pose information more efficiently, the positioning method is more favorable for deployment in the intelligent driving device, improves the safety in the automatic driving control process, and improves the safety of the automatic driving control process. Good to meet the needs of the field of autonomous driving.
  • an embodiment of the present disclosure further provides a positioning device, including: a first acquisition module, configured to acquire a target image obtained by collecting a target scene; and a first determination module, configured to determine the target image based on the target image.
  • the target object performs position matching;
  • the second determination module is configured to determine, based on the result of the position matching, the target pose information of the capture device that captures the target image.
  • an embodiment of the present disclosure further provides a driving control device for an intelligent driving device, including: a second acquisition module for acquiring video frame data collected by the intelligent driving device during driving; a detection module for using the first
  • the video frame data is processed to detect a target object in the video frame data; a control module is configured to control the target object based on the detected target object.
  • Intelligent driving device for an intelligent driving device, including: a second acquisition module for acquiring video frame data collected by the intelligent driving device during driving; a detection module for using the first
  • the video frame data is processed to detect a target object in the video frame data; a control module is configured to control the target object based on the detected target object.
  • Intelligent driving device for an intelligent driving device, including: a second acquisition module for acquiring video frame data collected by the intelligent driving device during driving; a detection module for using the first
  • the video frame data is processed to detect a target object in the video frame data; a control module is configured to control the target object based on the detected target object.
  • Intelligent driving device for
  • an optional implementation manner of the present disclosure further provides a computer device, a processor, and a memory, where the memory stores machine-readable instructions executable by the processor, and the processor is configured to execute the memory stored in the memory.
  • machine-readable instructions when the machine-readable instructions are executed by the processor, when the machine-readable instructions are executed by the processor, any one of the possible implementations of the first aspect or the second aspect is executed steps in .
  • an optional implementation manner of the present disclosure further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is run, executes any one of the first aspect or the second aspect above steps in a possible implementation.
  • the present disclosure provides a computer program, comprising computer-readable codes, when the computer-readable codes are executed in an electronic device, a processor in the electronic device executes the above-mentioned first aspect or the second aspect steps in any of the possible implementations.
  • FIG. 1 shows a flowchart of a positioning method provided by an embodiment of the present disclosure
  • FIG. 2 shows a flowchart of a specific method for determining, based on a target image, a first two-dimensional geometric feature of a first target object included in the target image in the target image provided by an embodiment of the present disclosure
  • FIG. 3 shows a schematic diagram of a semantic segmentation map obtained after semantic segmentation of a target image provided by an embodiment of the present disclosure
  • FIG. 4 shows a flowchart of a specific method for determining a first two-dimensional geometric feature of a first target object in a target image provided by an embodiment of the present disclosure
  • FIG. 5 shows a schematic diagram of determining a first two-dimensional geometric feature of a traffic light based on a semantic segmentation result of a traffic light provided by an embodiment of the present disclosure
  • FIG. 6 shows a flowchart of another specific method for determining a first two-dimensional geometric feature of a first target object in a target image provided by an embodiment of the present disclosure
  • FIG. 7 shows a schematic diagram of determining a first two-dimensional geometric feature of a straight-going pointing sign based on a semantic segmentation result of a straight-going pointing sign provided by an embodiment of the present disclosure
  • FIG. 8 shows a flowchart of another specific method for determining a first two-dimensional geometric feature of a first target object in a target image provided by an embodiment of the present disclosure
  • FIG. 9 shows a schematic diagram of determining a first two-dimensional geometric feature of a solid lane line based on a semantic segmentation result of the solid lane line provided by an embodiment of the present disclosure
  • FIG. 10 shows a schematic diagram of determining a correspondence between a first target object and a second target object according to an embodiment of the present disclosure
  • FIG. 11 shows a flowchart of a driving control method of an intelligent driving device provided by an embodiment of the present disclosure
  • FIG. 12 shows a schematic diagram of a positioning device provided by an embodiment of the present disclosure
  • FIG. 13 shows a schematic diagram of a driving control device of an intelligent driving device provided by an embodiment of the present disclosure
  • FIG. 14 shows a schematic diagram of a computer device provided by an embodiment of the present disclosure.
  • the research found that in the vision-based positioning method, the scene image obtained by shooting the target scene, and then extracting the feature points of the scene, and combining the extracted feature points with the features in the 3D scene model established in advance based on the target scene. Points are matched to obtain the pose information of the image acquisition device that acquires the scene image in the target scene.
  • the matching relationship based on feature points The process of solving the pose information takes a lot of time and the efficiency is low, so it cannot meet the needs of the autonomous driving field.
  • the present disclosure provides a positioning method, a driving control method, a device, a computer device and a storage medium, which utilize the geometric features of the target object to perform position matching on the first target object and the second target object, and based on the position matching As a result, the target pose information of the target image is obtained.
  • the matching process is simple, and the amount of data processed is less, so it has a faster processing speed and improves the efficiency of determining the target pose information of the acquisition device that collects the target image. .
  • the execution subject of the positioning method provided by the embodiment of the present disclosure is generally a computer device with a certain computing capability, such as a computer device.
  • terminal equipment can be user equipment (User Equipment, UE), mobile equipment, user terminal, terminal, cellular phone, cordless phone, Personal Digital Assistant (Personal Digital Assistant, PDA), handheld devices, computing devices, in-vehicle devices, wearable devices, etc.
  • the positioning method may be implemented by the processor invoking computer-readable instructions stored in the memory.
  • the method includes steps S101-S104, wherein:
  • S102 Based on the target image, determine the first two-dimensional geometric feature of the first target object included in the target image in the target image;
  • S103 Based on the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene, the first two-dimensional geometric feature of the first target object, and the correspondence between the first target object and the second target object , performing position matching on the first target object and the second target object;
  • S104 Based on the result of the position matching, determine the target pose information of the capture device that captures the target image.
  • the first target object in the target scene and the second target object in the target scene are used.
  • This process uses the geometric features of the target object to match the position of the first target object and the second target object, the matching process is simple, and the amount of data processed is less, so it has a faster processing speed and improves the determination of the target image to be collected.
  • the efficiency of collecting the target pose information of the device is simple, and the amount of data processed is less, so it has a faster processing speed and improves the determination of the target image to be collected.
  • the positioning method provided by the embodiments of the present disclosure can be applied to various fields, such as the field of automatic driving and the field of intelligent warehousing;
  • the target scene includes, for example, a road scene of automatic driving;
  • the target objects in the target scene may include road sign objects, for example including at least one of the following: street signs, traffic lights, zebra crossings, road directional signs, stop lines, solid lane lines, and dashed lane lines, and the like.
  • the target scene may also include a parking lot scene;
  • the target objects in the target scene include, for example, surveillance cameras, inductive card readers, vehicle detection lines, road pointing signs, parking line etc.
  • the target scene includes, for example, a warehouse; the target objects in the target scene include, for example, shelves in the warehouse, indicating landmarks, and the like.
  • the driving vehicle and the corresponding target scene may be determined according to the actual situation, which is not limited here.
  • the embodiments of the present disclosure take the target scene as a road on which an autonomous vehicle travels as an example, when acquiring a target image of the target scene, for example, an image acquisition device may be installed on the autonomous vehicle; the image acquisition device can scan the target scene in real time, Get the target image of the target scene.
  • other devices such as radar, depth camera, distance sensor, etc. may also be installed on the autonomous driving vehicle, and the autonomous driving vehicle may also perform positioning based on data obtained by other devices, and use the data provided based on the embodiments of the present disclosure.
  • the positioning result determined by the positioning method is combined with the positioning result determined by the detection data obtained by other equipment to obtain a more accurate positioning result.
  • the target image can be used to determine the first target object included in the target image and the first two-dimensional geometric feature of the first target object in the target image.
  • the first target object may include, for example, at least one of street signs, traffic lights, zebra crossings, road directional signs, prohibition marks, stop lines, solid lane lines and dashed lane lines in the target image; the above-mentioned target objects are relatively common in the target scene. iconic element. Since different target objects appear as different graphic features when shown in the target image, the corresponding two-dimensional geometric information can be determined according to the different target objects in the target image. in:
  • the graphic features of the first target object include, for example, having a rectangular outline.
  • the first target object is a first road sign object including a rectangular outline.
  • the first two-dimensional geometric feature of the first target object includes the vertices of the first target object.
  • the first target object is a zebra crossing
  • its outline can be shown by a plurality of continuous rectangles
  • the corresponding first two-dimensional geometric feature can include, for example, the vertex of one of the rectangles, or a plurality of continuously shown matrices corresponding vertices.
  • the graphic features of the first target object include, for example, having an irregular outline.
  • the first target object includes a second road sign object having an irregular outline.
  • the first two-dimensional geometric features of the first target object include contour lines and/or corner points of the first target object.
  • the first target object includes a right-turn directional sign in a road directional sign, its outline cannot be directly formed by basic outlines such as rectangles and circles, and its corresponding first two-dimensional geometric feature directly corresponds to its irregular outline.
  • the graphic features of the first target object include, for example, being shown in a line type.
  • the first target object is the third road sign including the line type.
  • the first two-dimensional geometric feature of the first target object includes: target line segments on the center lines of two adjacent first target objects.
  • the first target object when the first target object includes a solid lane line, because the solid lane line is long, only one end of the solid lane line may be shown in the target image, or neither end can be shown, and the corresponding first
  • the geometric feature corresponding to the two-dimensional geometric feature corresponds to a target line segment that belongs to the image area where the first target object is located and is located on the center line of the first target object.
  • the first two-dimensional geometric information of the first target object in the target image may, for example, be expressed as a two-dimensional coordinate value of the first target object in an image coordinate system corresponding to the target image.
  • the vertex of the first target object can be represented, for example, as the two-dimensional coordinate value of the vertex in the image coordinate system
  • the contour line can be represented, for example, as the two-dimensional coordinate value of the endpoint of the contour line in the image coordinate system
  • the target line segment It can be expressed as the two-dimensional coordinate value of the endpoint of the line segment in the image coordinate system.
  • the image coordinate system corresponding to the target image can be determined based on any pixel in the target image, for example, using any pixel as the coordinate origin to determine the image coordinate system; specifically, it can be based on the installation position of the image acquisition device in the autonomous vehicle Determine; for example, if the image acquisition device is installed at a higher position and the field of view is also higher, the image coordinate system can be determined by taking the lower pixel point in the target image as the origin. If the image acquisition device is installed at a lower position and the field of view is also lower, the image coordinate system can be determined by using the higher-position pixel point in the target image as the origin. In addition, the image coordinate system can also be established by taking the projected pixel point of the optical axis of the image acquisition device in the target image as the origin. The details can be determined according to the actual situation, and details are not repeated here.
  • a specific method for determining a first two-dimensional geometric feature of a first target object included in the target image in the target image based on the target image includes:
  • S201 Perform semantic segmentation processing on the target image, and determine the semantic segmentation results corresponding to multiple pixels in the target image respectively.
  • FCN Fully Convolutional Networks
  • CNN-CRF Convolutional Neural Networks-Fully Convolutional Networks
  • CDN encoder-decoder model
  • FPN feature pyramid model
  • FIG. 3 is a schematic diagram of a semantic segmentation map obtained by semantically segmenting a target image according to an embodiment of the present disclosure.
  • the area indicated by 31 represents traffic lights
  • the area indicated by 32 is the solid line of the road
  • the area indicated by 33 is the road directional sign
  • the area indicated by 34 is the broken line of the road.
  • S202 Determine a first two-dimensional geometric feature of the first target object in the target image based on the semantic segmentation results corresponding to the plurality of pixels and the respective positions of the plurality of pixels in the target image.
  • the corresponding method for determining the two-dimensional geometric feature of the first target object in the target image is also different.
  • the first two-dimensional geometric feature of the first target object in the target image can be determined in the following manner as shown in FIG. 4 :
  • S402 Based on the pixels belonging to the outline of the first target object, fit a bounding box corresponding to the first target object in the target image;
  • S403 Determine a first two-dimensional geometric feature of the first target object in the target image based on the vertices of the bounding box.
  • the pixels belonging to the outline of the first target object based on the semantic segmentation result for example, based on the semantic segmentation results of each pixel in the target image, it is determined from the target image that all pixels of the first target object are composed of pixels. area, and then determine the pixels at the edge of the area as pixels belonging to the outline of the first target object.
  • the contour line formed by the pixel points belonging to the contour of the first target object is usually a corrugated or jagged line with small fluctuations.
  • the edges of the contour line are not neat, it is difficult to identify the geometric features of the first target object based on the vertices determined based on the contour line; on the other hand, when the vertices determined based on the contour line represent the first target
  • the amount of data is also large, which is not conducive to matching between the same target objects. Therefore, in the embodiments of the present disclosure.
  • a bounding box corresponding to the first target object in the target image is obtained by fitting, and the vertices of the bounding box constitute the first two-dimensional geometric feature of the first target object.
  • the expression of the first two-dimensional geometric feature of the first target object is simplified to a certain extent, which is beneficial to the matching between the same target objects and reduces the difficulty of identification.
  • the bounding box corresponding to the first target object in the target image is obtained by fitting based on the pixels belonging to the contour of the first target object, for example, a plurality of straight lines may be determined based on the pixels belonging to the contour of the first target object, and A plurality of straight lines are fitted to obtain the bounding box of the first target object.
  • the obtained bounding box is approximately a rectangle.
  • the contour lines formed by the pixels belonging to the contour of the first target object may partially overlap. , or surround the area where the first target object is located.
  • the bounding box can surround the pixels corresponding to the first target object in the target image, and the specific first two-dimensional geometric features of the first target object in the target image can be more accurately represented by the vertices of the bounding box.
  • the coordinate values in the image coordinate system corresponding to the target image can represent the specific position of the first target object in the target image.
  • the position of the rectangle in the target image can be determined according to the two-dimensional coordinate values of the opposite corners in the rectangle.
  • the upper left corner vertex and the lower right corner vertex of the bounding box, or The two-dimensional coordinate values of the lower left corner vertex and the lower right corner vertex of the frame in the target image are taken as the specific position of the first target object in the target image.
  • Using this method can ensure that the geometric features of the first target object can be accurately expressed, and at the same time reduce the amount of data in subsequent processing processes (eg, performing position matching between the first target object and the second target object).
  • the two-dimensional coordinate values of the four top corners of the rectangular frame may also be directly determined as the representation of the specific position of the first target object in the target image.
  • the specific position of the first target object in the target image obtained by processing with this method makes the first two-dimensional geometric feature of the target object more readable.
  • FIG. 5 a schematic diagram of determining a first two-dimensional geometric feature of a traffic light based on a semantic segmentation result of a traffic light provided by an embodiment of the present disclosure; wherein, a in FIG. 5 indicates that when the first target object includes a traffic light, the first A schematic diagram of the semantic segmentation result of the target object in the target image; in FIG. 5 b represents a schematic diagram of a bounding box corresponding to the first target object, and 51 and 52 represent two vertices corresponding to the bounding box.
  • the first two-dimensional geometric feature of the first target object in the target image can be determined in the following manner as shown in FIG. 6 :
  • S601 Based on the semantic segmentation result, determine the pixel points belonging to the contour of the first target object from the target image;
  • S602 Based on the positions of the pixels belonging to the outline of the first target object in the target image, obtain the outline of the first target object;
  • S603 Determine a first two-dimensional geometric feature of the first target object in the target image based on the outline of the first target object.
  • the first target object includes a road surface pointing mark
  • the shape of the first target object is irregular, when determining the first two-dimensional geometric feature of the first target object, it is necessary to The pixel points of the outline of the object determine the first two-dimensional geometric feature of the first target object.
  • the vertex used for determining the first two-dimensional geometric feature can be selected, for example, as a turning point that turns at a larger angle in the edge surrounding the first target object. Since the position of the line corresponding to the turning point in the actual outline of the first target object fluctuates greatly, the change of the turning point, etc. is large, so even if the outline of the first target object fluctuates slightly, the interference to the vertex recognition is small, so that the It is easier to identify the contour line determined based on the semantic segmentation result to determine the vertex, that is, the specific position of the first target object in the target image can be easily determined.
  • FIG. 7 is a schematic diagram of determining a first two-dimensional geometric feature of a straight-going pointing sign based on a semantic segmentation result of a straight-going pointing sign provided by an embodiment of the present disclosure.
  • FIG. 7 a shows a schematic diagram of a semantic segmentation map corresponding to a straight-line pointing mark
  • FIG. 7 b shows a schematic diagram of a contour line of a first target object.
  • the contour lines are irregular arrow-shaped.
  • 71 , 72 and 73 in b in FIG. 7 respectively represent a plurality of vertices obtained by identification to determine the contour line corresponding to the first target object in the target image.
  • the first two-dimensional geometric feature of the first target object in the target image can be determined in the following manner as shown in FIG. 8 :
  • S802 Based on the two-dimensional coordinate values in the target image of the pixels located on the center line and belonging to the image area where the first target object is located, determine the target line segment that belongs to the image area where the first target object is located and is located on the center line;
  • S803 Obtain a first two-dimensional geometric feature of the first target object based on the target line segment.
  • the first target object includes a target object whose at least one end is not shown in the target image in the outline
  • the stop line and the solid lane line are on the road during the normal driving of the vehicle, so the target line segment of the stop line and the solid lane line is selected as the first target object.
  • the target line segment can be expressed as the two-dimensional coordinate value of the endpoint of the target line segment in the target image.
  • the center line of the first target object may be determined first.
  • at least one of the following methods can be used: a method based on topology refinement, a method based on distance transformation, a method based on path planning, and a method based on tracking.
  • the semantic segmentation map corresponding to the first target object may be determined first, and then the boundary of the semantic segmentation map is iteratively processed by using the morphological principle, until the boundary of the semantic segmentation map is eroded and eliminated.
  • the center line corresponding to the first target object in the semantic segmentation map is obtained. Since different centerline extraction methods are applicable to different scenarios and have different image quality requirements, the selection of a specific centerline extraction method and a specific execution process can be determined according to the actual situation, which will not be repeated here.
  • two points may also be determined on the centerline as endpoints of the target line segment.
  • a point-by-point search method can be used to determine whether the points on the centerline have corresponding pixels in the image area corresponding to the first target object one by one; Pixels located on the center line are determined in the image area corresponding to a target object. At this point, all pixels belonging to the image area where the first target object is located and located on the center line can be determined, and the two farthest pixels are used as endpoints of the target line segment. Then, the target line segment determined by the two farthest pixel points is used as the first two-dimensional geometric feature of the first target object.
  • FIG. 9 is a schematic diagram of determining a first two-dimensional geometric feature of a solid lane line based on a semantic segmentation result of a solid lane line provided by an embodiment of the present disclosure.
  • a in Figure 9 represents the semantic segmentation map of the solid lane line
  • b in Figure 9 represents the target line segment corresponding to the solid lane line.
  • the first two-dimensional geometric feature of the first target object can be determined.
  • the first target object includes at least one of the first road sign object and the second road sign object
  • the first two-dimensional geometric feature of the first target object can be represented as p j , for example; where j represents The jth first target object among the plurality of first target objects.
  • the first target object includes at least one of the third road sign objects
  • the first two-dimensional geometric feature of the first target object can be represented as l i ; first target object.
  • the second target object corresponds to the first target object may include at least one of street signs, traffic lights, zebra crossings, solid lane lines, road directional signs, prohibition marks, stop lines, and dashed lane lines.
  • the scene coordinate system corresponding to the target scene where the second target object is located may include, for example, a pre-established scene coordinate system.
  • the scene coordinate system is a three-dimensional coordinate system established for the target scene. Specifically, the world coordinate system may be directly selected as the scene coordinate system, or any position point in the target scene may be used as the origin to establish the scene coordinate system.
  • the specific scene coordinate system can be determined according to the actual situation, and will not be repeated here.
  • the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene can be determined.
  • the three-dimensional geometric feature of the second target object can be predetermined, for example, for example, can use Simultaneous Localization and Mapping (Simultaneous Localization and Mapping, SLAM) modeling, motion recovery structure (Structure-From-Motion, SFM) modeling to determine the three-dimensional geometric features of the second target object in the target scene.
  • Simultaneous Localization and Mapping Simultaneous Localization and Mapping, SLAM
  • motion recovery structure Structure-From-Motion
  • the three-dimensional geometric feature of the second target object includes, for example, each vertex in the second target object.
  • the corresponding three-dimensional geometric feature may also include, for example, corner points of the second target object.
  • the three-dimensional geometric feature of the second target object may be represented by the three-dimensional coordinate value of each vertex in the three-dimensional coordinate system.
  • the three-dimensional geometric feature of the second target object includes, for example, a line segment located on the center line of the second target object and belonging to the second target object.
  • the three-dimensional geometric feature of the second target object may be represented by the three-dimensional coordinate value of the end point of the line segment in the three-dimensional coordinate system.
  • the three-dimensional geometric feature of the second target object can be represented as P, for example j ; wherein, j represents the jth second target object among the plurality of second target objects.
  • the three-dimensional geometric feature of the second target object may be represented as L i , for example; The ith second target object among the two target objects.
  • the corresponding relationship between the first target object and the second target object may also be generated based on the three-dimensional geometric feature of the second target object in the target scene and the first two-dimensional geometric feature of the first target object.
  • Embodiments of the present disclosure also provide a specific method for generating a correspondence between a first target object and a second target object, including: based on the three-dimensional geometric features of the second target object in the target scene in the target scene, the The first two-dimensional geometric feature generates a corresponding relationship between the first target object and the second target object.
  • the second target object includes: in the case of at least one of the target object whose outline is shown by at least one rectangle or the target object shown by using irregular graphics, the first target object and the first target object generated in the following manner can be used.
  • the first target object and the second target object are matched to obtain the first target object.
  • the pose when acquiring the initial pose information of the acquisition device that collects the target image, since the autonomous vehicle normally drives on the road in the target scene, the pose usually does not change significantly, so for example, it can be Obtain the previously determined pose information of the autonomous vehicle on the road in advance as the initial pose information; or, since the information of the road in the target scene can be easily obtained, for example, a global positioning system (Global Positioning System, GPS) is used to determine The position information of the road is then estimated based on the position information of the road in the target scene to obtain the initial pose information of the autonomous vehicle.
  • the initial pose information can be represented as T 0 , for example.
  • the specific method for determining the information of the initial pose of the target image can be determined according to the actual situation, and details are not described herein again.
  • the pose information of the acquisition device that captures the target image is associated with the pose information corresponding to the autonomous vehicle.
  • the pose of the autonomous vehicle can be determined based on the relative pose relationship between the image acquisition device and the autonomous vehicle
  • the pose information of the autonomous driving vehicle can be determined based on the pose information of the collection device of the collected target image.
  • the second target After determining the initial pose information of the capture device that captures the target image and the three-dimensional geometric features of the second target object in the target scene, the second target can be obtained by projecting the second target object into the image coordinate system of the target image. The first projected geometric feature of the object in the image coordinate system.
  • the following methods can be used: The three-dimensional geometric features of the second target object are converted from the scene coordinate system to the world coordinate system, and the model space conversion from scene space coordinates to world space coordinates is completed; The three-dimensional geometric features in the coordinate system are converted from the world coordinate system to the image coordinate system, and the observation space conversion from the world space coordinates to the camera space coordinates is completed.
  • the first projected geometric feature of the second target object in the image coordinate system can be obtained.
  • the specific method for determining the first projected geometric feature of the second target object may be determined according to the actual situation, and details are not described herein again.
  • the obtained first projected geometric feature of the second target object in the image coordinate system and the first target object there is also a correspondence between the first two-dimensional geometric features in the image coordinate system, so that it can be based on the first two-dimensional geometric features of the first target object in the image coordinate system and the second target object in the image coordinate system.
  • the first projected geometric feature matches the first target object and the second target object to obtain the corresponding relationship between the first target object and the second target object.
  • KNN k-Nearest Neighbor
  • the second target object includes at least one of the target object whose outline is shown by at least one rectangle or the target object shown by using irregular graphics
  • the corresponding relationship between the first target object and the second target object can be determined.
  • the specific method for generating the correspondence between the first target object and the second target object includes: :
  • the first target object in the target image is projected into the target plane to obtain the second projected geometric feature of the first target object in the target plane;
  • the second projected geometric feature of the first target object in the target plane and the geometric feature of the second target object in the target plane, the first target object and the second target object are matched to obtain the first target object and the second target object correspondence.
  • the geometric features of the second target object in the target plane are determined based on the three-dimensional geometric features of the second target object in the scene coordinate system.
  • a homography matrix (homography) between the target image and the target plane where the second target object is located is used to project the first target object in the target image onto the target plane.
  • the coordinate value of any pixel of the second target object on the target plane can be obtained, wherein the second target object
  • the coordinate system O 2 (x 2 , y 2 , z 2 ) of the corresponding positions of the at least two pixel points on the target image, and then, for example, the homography matrix can be determined according to the following formula (1), for example, it can be represented by H:
  • the first target object may be projected onto the target plane, so as to obtain the second projected geometric feature of the first target object in the target plane.
  • the first target object Matching with the second target object to obtain the corresponding relationship between the first target object and the second target object.
  • the features characterizing the vertices of the first target object in the second projected geometric feature and the second target object in the target plane are utilized The geometric features representing the vertices determined by the second target object are matched to obtain the corresponding relationship between the first target object and the second target object.
  • the first target object includes the first road sign with a rectangular outline
  • the first target object is represented by the features of the vertexes of the first target object in the second projected geometric features and the geometric features of the second target object in the target plane.
  • the features of the vertices determined by the two target objects are matched, that is, the correspondence between the first target object and the second target object can be relatively simply determined.
  • the features representing the contour and/or corner of the first target object in the second projected geometric feature, and the second target The objects are matched with the geometric features in the target plane representing the contour lines and/or corner points determined by the second target object to obtain the correspondence between the first target object and the second target object.
  • the features representing the contour lines and/or corners of the first target object and the second target object are correspondingly included in the second projected geometric features.
  • the way of matching the features representing the contour lines and/or corner points determined by the second target object among the geometric features in the target plane may be similar to the matching mode corresponding to the first road sign, using the contour line vertices in the contour line and the / or directly use the determined corner points for matching. In this way, for the second road sign object whose graphical expression is more complex than that of the first road sign object, the matching accuracy can be correspondingly improved.
  • the first target object includes a line-type third road sign object
  • the corresponding relationship of the target object In the case where the first target object includes a line-type third road sign object, perform maximum graph matching on the second projected geometric feature and the geometric feature of the second target object in the target plane to obtain the first target object and the second target object. The corresponding relationship of the target object.
  • a plurality of candidate matching pairs can be randomly constructed in any matching manner, and each candidate The matching pair includes a first target object and a second target object, and each first target object is included in only one candidate matching pair, and each second object is included in only one candidate matching pair.
  • the straight-line distance between the first target object and the second target object in each candidate matching pair may be calculated, and the average value of the above-mentioned straight-line distance corresponding to each candidate matching pair may be determined.
  • the above-mentioned matching mode with an excessively large average value is removed, thereby obtaining the matching relationship between the first target object and the second target object.
  • the matching process can be completed even when the number of the first target objects to be matched is different from the number of the second target objects, and the matching between a smaller number of first target objects and a larger number of second target objects can be obtained.
  • the automatic identification of the missing line-shaped third road sign in the target image is realized through maximum graph matching, and a relatively accurate matching result can be obtained.
  • the first target object since the first target object includes a line-type third road sign object, it cannot show the information of all the points it includes, especially all the points representing its actual position, after projection. Therefore, the above method of matching the first road sign object and/or the second road sign object using at least one of points, contour lines and corner points is not applicable.
  • the second projected geometric feature and the geometric feature of the second target object in the target plane can be directly matched without re-collecting the third road sign object to ensure the efficiency of positioning .
  • the nearest neighbor matching method is used.
  • the obtained correspondence is inaccurate, so that missed detection lines are likely to occur during matching, so that the determined correspondence between the first target object and the second target object is inaccurate.
  • an optimal matching algorithm Kuhn-Munkras, KM
  • Kuhn-Munkras, KM an optimal matching algorithm
  • the weight value of the first target object corresponding to each target line segment can remove the matches whose average distance to the target line segment is too large in the matching candidates, so as to obtain the correspondence between the first target object and the second target object.
  • FIG. 10 is a schematic diagram of determining a correspondence between a first target object and a second target object according to an embodiment of the present disclosure.
  • 11 represents the first target object
  • 12 represents the second target object
  • 13 represents the first target object that does not have a matching second target object after matching the lines through the maximum graph matching algorithm, that is, 13 represents Lines that are missed in the target image
  • 14 represent a set of corresponding first target objects and second target objects.
  • the missing detection lines may not be processed.
  • the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene the first target object's third A two-dimensional geometric feature for performing position matching on the first target object and the second target object.
  • the three-dimensional geometric feature of the second target object in the scene coordinate system is matched with the first target object.
  • the matching loss caused by the deviation between the initial pose information and the actual pose information can be determined, so as to determine the target pose information of the acquisition device that collects the target image based on the determined matching loss.
  • the second target object Based on the initial pose information of the capture device that captures the target image and the three-dimensional geometric features of the second target object in the target scene in the scene coordinate system corresponding to the target scene, project the second target object into the image coordinate system to obtain the first target object.
  • the third projected geometric feature of the target object in the image coordinate system Based on the third projected geometric feature of the second target object in the image coordinate system, the first two-dimensional geometric feature of the first target object, and the corresponding relationship between the first target object and the second The target object and the second target object perform position matching.
  • the method for determining the third projected geometric feature of the second target object in the image coordinate system is similar to the above-mentioned method for determining the first projected geometric feature, and details are not described herein again.
  • the determined third projected geometric feature for example, can be expressed as ⁇ (L i ,T 0 ); wherein, ⁇ is a projection function, which is used to project the three-dimensional geometric feature Li of the second target object into the image coordinate system according to the initial pose information T 0 .
  • the determined third projected geometric feature can be expressed as ⁇ (P j , T 0 ); wherein, the projection function ⁇ is used to project the three-dimensional geometric feature P j of the second target object into the image coordinate system according to the initial pose information T 0 .
  • the projection vertices or projection target line segments in the second target object correspond to vertices or target line segments in the first target object.
  • the specific method for determining the projection vertex or the projection target line segment is similar to the above-mentioned method for determining the vertex or the target line segment in the first target object, and will not be repeated here.
  • the first target object includes at least one of the target objects whose at least one end is not shown in the target image in the outline
  • the first target The first two-dimensional geometric feature of the object is positionally matched with the third projected geometric feature of the corresponding second target object in the image coordinate system, so as to determine the correspondence between the first target object and the second target object.
  • the first two-dimensional geometric feature of the first target object is determined by the coordinate value of the end point of the target line segment of the first target object, when the corresponding relationship is determined, only the position matching of the end point needs to be performed, and the amount of calculation is less. , which makes it more efficient when matching positions.
  • a specific method for performing position matching on a first target object and a second target object includes: A first two-dimensional geometric feature of a target object is subjected to interpolation processing to obtain a second two-dimensional geometric feature of the first target object; wherein the second two-dimensional geometric feature includes: coordinate values of multiple vertices in the target image, and multiple The coordinate values of the interpolation points in the target image; based on the second two-dimensional geometric features, three-dimensional geometric features, and the corresponding relationship between the first target object and the second target object, perform point-to-point positions on the first target object and the second target object. match.
  • the first two-dimensional geometric feature of the first target object may be, for example, based on the first target object.
  • the coordinate values of two or four vertices are obtained, and there may be fewer vertices.
  • the determined multiple interpolation points can form the sparse outline of the first target object, so that the two
  • the difference in the number of vertices in the direction of each coordinate axis is small, so as to balance the weights between different semantics, and at the same time, it can also alleviate the problem of poor matching when using fewer vertices for position matching processing.
  • At least one of the following methods can be used: Taylor Interpolation, Lagrange Interpolation, Newton Interpolation, and Hermitian Interpolation ( Hermite Interpolation).
  • Taylor Interpolation Lagrange Interpolation
  • Newton Interpolation Newton Interpolation
  • Hermitian Interpolation Hermite Interpolation
  • the specific interpolation method can be selected according to the actual situation, and will not be repeated here.
  • the position matching error may be determined based on the position matching result, and based on the position matching error and the position matching error of the capture device that captures the target image
  • the initial pose information determines the target pose information of the acquisition device that collects the target image.
  • the corresponding position matching error can be determined.
  • the position matching error corresponding to the third projected geometric feature ⁇ (L i , T 0 ) can be expressed as, for example, D l ( ⁇ (L i , T 0 ), l i ), where D l represents the end point of the projected target line segment to the target The residual term for the distance to the centerline where the line segment lies.
  • the position matching error corresponding to the third projected geometric feature ⁇ (P j , T 0 ) can be expressed as, for example, D p ( ⁇ (P j , T 0 ), p i ), where D p represents the reprojection between the vertex and the projected vertex error.
  • Q represents the total number of target objects whose contours included in the first target object are represented by at least one rectangle or are represented by irregular graphics; P represents that at least one end of the contours included in the first target object is not in the The total number of target objects shown in the target image; error represents the determined matching loss.
  • the following methods may be used: detecting whether a preset iteration stop condition is met; if the iteration stop condition is met , the initial pose information obtained in the last iteration is determined as the target pose information; if the iteration stop condition is not met, a new initial pose information is determined based on the position matching error and the initial pose information in the latest iteration process pose information, and return to the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene, the first two-dimensional geometric feature of the first target object, and the first target object and the second target object.
  • the correspondence between the target objects, and the step of performing position matching on the first target object and the second target object may be used: detecting whether a preset iteration stop condition is met; if the iteration stop condition is met , the initial pose information obtained in the last iteration is determined as the target pose information; if the iteration stop condition is not met, a new initial pose information is determined based on the position matching error and the
  • the iteration stop condition includes at least one of the following: the number of iterations is greater than a preset threshold of the number of iterations; the position matching error between the first target object and the second target object is less than a preset loss threshold.
  • the preset number of iterations threshold such as 6 or 8 times, may be determined based on experience, so that the matching loss after enough iterations is small.
  • a smaller loss threshold can be set to make the obtained target pose information more confident .
  • the specific selection of the iterative stop condition can be determined according to the actual situation, and details are not repeated here.
  • the direction of the iteration is determined as the direction of reducing the position matching error
  • the initial pose information in the latest iteration process is determined as the new initial position
  • the initial pose information at this time can be determined as the target pose information, for example, it can be expressed as T aim .
  • the target pose information T aim that is, the target pose information T aim of the target image can be determined.
  • an embodiment of the present disclosure also provides a driving control method for an intelligent driving device.
  • the driving control method for an intelligent driving device includes steps S1101 to S1103 , wherein:
  • S1101 Acquire video frame data collected by the intelligent driving device during driving
  • S1102 Obtain a target detection neural network by using the positioning method provided by the embodiment of the present disclosure, and detect the target object in the video frame data;
  • the driving device is, for example, but not limited to, any one of the following: an autonomous vehicle, a vehicle equipped with an advanced driving assistance system (Advanced Driving Assistance System, ADAS), or a robot, and the like.
  • an autonomous vehicle a vehicle equipped with an advanced driving assistance system (Advanced Driving Assistance System, ADAS), or a robot, and the like.
  • ADAS Advanced Driving Assistance System
  • robot a robot, and the like.
  • Controlling the traveling device includes, for example, controlling the traveling device to accelerate, decelerate, turn, and brake, or play voice prompt information to prompt the driver to control the traveling device to accelerate, decelerate, turn, and brake.
  • the positioning method provided by the embodiment of the present disclosure can be used to determine the target pose information more efficiently, the positioning method is more conducive to deployment in the intelligent driving device, improves the safety in the automatic driving control process, and better Meet the needs of autonomous driving.
  • the writing order of each step does not mean a strict execution order but constitutes any limitation on the implementation process, and the specific execution order of each step should be based on its function and possible Internal logic is determined.
  • the embodiment of the present disclosure also provides a positioning device corresponding to the positioning method. Since the principle of solving the problem of the device in the embodiment of the present disclosure is similar to the above-mentioned positioning method in the embodiment of the present disclosure, the implementation of the device can refer to the method of implementation, and the repetition will not be repeated.
  • the device includes: a first acquisition module 121, a first determination module 122, a matching module 123, and a second determination module 124; wherein,
  • the first acquisition module 121 is used to acquire the target image obtained by collecting the target scene; the first determination module 122 is used to determine, based on the target image, that the first target object included in the target image is in the target image The first two-dimensional geometric feature in A two-dimensional geometric feature, and the corresponding relationship between the first target object and the second target object, perform position matching on the first target object and the second target object; the second determining module 124 is used for Based on the result of the position matching, the target pose information of the capture device that captures the target image is determined.
  • the first two-dimensional geometric feature of the first target object when the first target object includes a first road sign object with a rectangular outline, the first two-dimensional geometric feature of the first target object includes: the first the vertex of the target object; in the case where the first target object includes a second road sign object with an irregular contour, the first two-dimensional geometric feature of the first target object includes: the contour of the first target object Lines and/or corners; in the case that the first target object includes a line-type third road sign object, the first two-dimensional geometric features of the first target object include: belonging to the first target object The target line segment located in the image area and located on the center line of the first target object.
  • the first determination module 122 determines, based on the target image, the first two-dimensional geometric feature of the first target object included in the target image in the target image, It is used for: performing semantic segmentation processing on the target image, and determining the semantic segmentation results corresponding to multiple pixels in the target image; based on the semantic segmentation results corresponding to the multiple pixels, and the At the position in the target image, a first two-dimensional geometric feature of the first target object in the target image is determined.
  • the first determination module 122 is based on the semantic segmentation results corresponding to the plurality of pixel points, and the respective positions of the plurality of pixel points in the target image, when determining the first two-dimensional geometric feature of the first target object in the target image, for: based on the semantic segmentation result, from The pixel points belonging to the outline of the first target object are determined in the target image; based on the pixel points belonging to the outline of the first target object, the corresponding pixel points of the first target object in the target image are obtained by fitting.
  • a bounding box determining a first two-dimensional geometric feature of the first target object in the target image based on the vertices of the bounding box.
  • the first determination module 122 is based on the semantic segmentation results corresponding to the plurality of pixel points. , and the respective positions of the plurality of pixels in the target image, when determining the first two-dimensional geometric feature of the first target object in the target image, for: based on the semantic segmentation result, Determine the pixel points belonging to the outline of the first target object from the target image; based on the positions of the pixels belonging to the outline of the first target object in the target image, obtain the pixel points of the first target object. Outline; based on the outline of the first target object, determine a first two-dimensional geometric feature of the first target object in the target image.
  • the first determination module 122 is based on the semantic segmentation results corresponding to the plurality of pixel points, and The respective positions of the plurality of pixel points in the target image, when determining the first two-dimensional geometric feature of the first target object in the target image, are used for: based on the semantic segmentation result, fitting Obtain the center line of the first target object; based on the two-dimensional coordinate values in the target image of the pixels located on the center line and belonging to the image area where the first target object is located, determine that the first object belongs to the first target The target line segment in the image area where the object is located and on the center line; the first two-dimensional geometric feature of the first target object is obtained based on the target line segment.
  • a generation module 125 is also included, configured to: based on the three-dimensional geometric features of the second target object in the target scene in the target scene, the first and second A dimensional geometric feature is used to generate the corresponding relationship between the first target object and the second target object.
  • the generation module 125 is based on the three-dimensional geometric feature of the second target object in the target scene and the first two-dimensional geometric feature of the first target object. , when the corresponding relationship between the first target object and the second target object is generated, it is used for: based on the initial pose information of the capture device that captures the target image, and the second target object in the For the three-dimensional geometric feature in the target scene, project the second target object into the image coordinate system of the target image to obtain the first projected geometric feature of the second target object in the image coordinate system; Based on the first two-dimensional geometric feature of the first target object in the image coordinate system and the first projected geometric feature of the second target object in the image coordinate system, the first target object is Matching with the second target object to obtain the corresponding relationship between the first target object and the second target object.
  • the generation module 125 is based on the three-dimensional geometric feature of the second target object in the target scene and the first two-dimensional geometric feature of the first target object.
  • the generation module 125 when generating the corresponding relationship between the first target object and the second target object, for: based on the homography matrix between the target image and the target plane where the second target object is located, the The first target object in the target image is projected into the target plane to obtain the second projected geometric feature of the first target object in the target plane; based on the projection of the first target object in the target plane The second projected geometric feature and the geometric feature of the second target object in the target plane are matched to the first target object and the second target object to obtain the first target object and the second target object.
  • the corresponding relationship of the second target object; wherein, the geometric feature of the second target object in the target plane is determined based on the three-dimensional geometric feature of the second target object in the scene coordinate system.
  • the generating module 125 is based on the second projected geometric feature of the first target object in the target plane and the geometry of the second target object in the target plane. feature, when the first target object and the second target object are matched to obtain the corresponding relationship between the first target object and the second target object, it is used for: when the first target object includes a In the case of a first road sign object with a rectangular outline, the features representing the vertices of the first target object in the second projected geometric features and the geometric features of the second target object in the target plane are used Matching the features of the determined vertices representing the second target object to obtain the correspondence between the first target object and the second target object; where the first target object includes a second road sign with an irregular outline In the case of an object, use the second projected geometric features to characterize the contours and/or corners of the first target object, and the second target object to be represented by the geometric features in the target plane.
  • the second determination module 124 when determining the target pose information of the capture device that captures the target image based on the result of the position matching, is used to: determine the position based on the result of the position matching. Matching error; based on the position matching error and the initial pose information of the capture device that captures the target image, determine the target pose information of the capture device that captures the target image.
  • the second determination module 124 determines the target position of the capture device that captures the target image based on the position matching error and the initial pose information of the capture device that captures the target image.
  • the pose information it is used to: detect whether a preset iteration stop condition is met; if the iteration stop condition is met, determine the initial pose information obtained by the last iteration as the target pose information; In the case where the iteration stop condition is not satisfied, based on the position matching error and the initial pose information in the latest iteration process, determine new initial pose information, and return to the first position based on the target scene.
  • the iteration stop condition includes any one of the following: the number of iterations is greater than a preset number of iterations threshold; the position matching error between the first target object and the second target object is less than a preset value. loss threshold.
  • the matching module 123 is based on the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene, the first target object A two-dimensional geometric feature and the corresponding relationship between the first target object and the second target object, when performing position matching on the first target object and the second target object, are used for: in the first target object and the second target object.
  • a target object When a target object includes a second road sign object with an irregular contour, perform interpolation processing on the first two-dimensional geometric feature of the first target object to obtain the second two-dimensional geometric feature of the first target object ; wherein, the second two-dimensional geometric feature includes: a plurality of vertices and a plurality of interpolation points; based on the second two-dimensional geometric feature, the three-dimensional geometric feature and the first target object and the second The correspondence between the target objects, and the point-to-point position matching is performed on the first target object and the second target object.
  • the matching module 123 is based on the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene, the first target object A two-dimensional geometric feature and the corresponding relationship between the first target object and the second target object, when performing position matching on the first target object and the second target object, are used to: based on the acquisition of the The initial pose information of the acquisition device of the target image and the three-dimensional geometric feature of the second target object in the target scene in the scene coordinate system corresponding to the target scene, and the second target object is projected to the target In the image coordinate system of the image, the third projected geometric feature of the second target object in the image coordinate system is obtained; based on the third projected geometric feature of the second target object in the image coordinate system .
  • the first two-dimensional geometric feature of the first target object is to perform position matching on the first target object and the second target object having a corresponding relationship.
  • the embodiment of the present disclosure also provides a driving control device of an intelligent driving device corresponding to the driving control of the intelligent driving device, because the principle of solving the problem of the device in the embodiment of the present disclosure is the same as the above-mentioned positioning method of the embodiment of the present disclosure. Similar, therefore, the implementation of the apparatus may refer to the implementation of the method, and repeated descriptions will not be repeated.
  • FIG. 13 is a schematic diagram of a driving control device of an intelligent driving device according to an embodiment of the present disclosure
  • the device includes: a second acquisition module 131 , a detection module 132 , and a control module 133 ; wherein,
  • the second acquisition module 131 is configured to acquire video frame data collected by the intelligent driving device during driving;
  • a detection module 132 configured to process the video frame data by using any of the positioning methods provided by the embodiments of the present disclosure, and detect a target object in the video frame data;
  • the control module 133 is configured to control the intelligent driving device based on the detected target object.
  • An embodiment of the present disclosure further provides a computer device.
  • the schematic structural diagram of the computer device provided by the embodiment of the present disclosure includes:
  • the processor 141 performs the following steps:
  • Acquiring video frame data collected by an intelligent driving device during driving processing the video frame data by using any positioning method provided by the embodiments of the present disclosure, and detecting a target object in the video frame data; based on the detected target object , to control the intelligent driving device.
  • the above-mentioned memory 142 includes a memory 1421 and an external memory 1422; the memory 1421 here is also called internal memory, which is used to temporarily store the operation data in the processor 141 and the data exchanged with the external memory 1422 such as the hard disk.
  • the external memory 1422 performs data exchange.
  • Embodiments of the present disclosure further provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium.
  • a computer program is stored on the computer-readable storage medium.
  • the storage medium may be a volatile or non-volatile computer-readable storage medium.
  • Embodiments of the present disclosure further provide a computer program product, where the computer program product carries program code, and the program code includes instructions that can be used to execute the positioning method described in the above method embodiments, or the driving control method for an intelligent driving device
  • the computer program product carries program code
  • the program code includes instructions that can be used to execute the positioning method described in the above method embodiments, or the driving control method for an intelligent driving device
  • the above-mentioned computer program product can be specifically implemented by means of hardware, software or a combination thereof.
  • the computer program product is embodied as a computer storage medium, and in another optional embodiment, the computer program product is embodied as a software product, such as a software development kit (Software Development Kit, SDK), etc. Wait.
  • the units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.
  • each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
  • the functions, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a processor-executable non-volatile computer-readable storage medium.
  • the computer software products are stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program codes .

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Geometry (AREA)
  • Image Analysis (AREA)

Abstract

一种定位方法、行驶控制方法、装置、计算机设备及存储介质，其中，该方法包括获取对目标场景进行采集得到的目标图像（S101）；基于目标图像，确定目标图像中包括的第一目标对象在目标图像中的第一二维几何特征（S102）；基于目标场景中的第二目标对象在目标场景对应的场景坐标系中的三维几何特征、第一目标对象的第一二维几何特征、以及第一目标对象和第二目标对象的对应关系，对第一目标对象和第二目标对象进行位置匹配（S103）；基于位置匹配的结果，确定采集目标图像的采集设备的目标位姿信息（S104）。

Description

定位方法、行驶控制方法、装置、计算机设备及存储介质
本公开要求在2021年03月31日提交中国专利局、申请号为202110349351.3、申请名称为“定位方法、行驶控制方法、装置、计算机设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
技术领域
本公开涉及计算机视觉技术领域，具体而言，涉及一种定位方法、行驶控制方法、装置、计算机设备及存储介质。
背景技术
在自动驾驶领域中，确定车辆的位姿信息是非常基础的环节，在自动驾驶中的车辆感知、决策以及路径规划等任务，都需要在确定当前车辆位姿信息的基础上进一步处理实现，因此确定的位姿信息的精度直接影响后续感知的准确率、决策的合理性以及规划的精细程度。
发明内容
本公开实施例至少提供一种定位方法、行驶控制方法、装置、计算机设备及存储介质。
第一方面，本公开实施例提供了一种定位方法，包括：获取对目标场景进行采集得到的目标图像；基于所述目标图像，确定所述目标图像中包括的第一目标对象在所述目标图像中的第一二维几何特征；基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配；基于位置匹配的结果，确定采集所述目标图像的采集设备的目标位姿信息。
该实施方式中，通过利用目标对象的几何特征，对第一目标对象和第二目标对象进行位置匹配，匹配过程简单，且处理的数据量更少，因此具有更快的处理速度，可以提升确定目标图像的采集设备的目标位姿信息时的效率。
一种可选的实施方式中，在所述第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，所述第一目标对象的第一二维几何特征，包括：所述第一目标对象的顶点；在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，所述第一目标对象的第一二维几何特征包括：所述第一目标对象的轮廓线和/或角点；在所述第一目标对象包括线型的第三道路标志对象的情况下，所述第一目标对象的第一二维几何特征，包括：属于所述第一目标对象所在图像区域、且位于所述第一目标对象的中心线上的目标线段。
该实施方式中，通过对不同的第一目标对象有针对性的确定对应的第一二维几何特征，能够使得不同的第一目标对象的第一二维几何特征容易识别得到，并且可以较为准确的表征第一目标对象在目标场景中的实际位置。
一种可选的实施方式中，所述基于所述目标图像，确定所述目标图像中包括的第一目标对象在所述目标图像中的第一二维几何特征，包括：对所述目标图像进行语义分割处理，确定所述目标图像中多个像素点分别对应的语义分割结果；基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征。
该实施方式中，由于在对目标图像进行语义分割处理时，可以较为准确地确定目标图像中多个像素点分别对应的语义分割结果，因此利用多个像素点分别对应的语义分割结果确定多个像素点在目标图像中的位置时，得到的第一二维几何特征更为准确。
一种可选的实施方式中，在所述第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，所述基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征，包括：基于所述语义分割结果，从所述目标图像中确定属于所述第一目标对象的轮廓的像素点；基于属于所述第一目标对象的轮廓的像素点，拟合得到所述第一目标对象在所述目标图像中对应的包围框；基于所述包围框的顶点确定所述第一目标对象在所述目标图像中的第一二维几何特征。
一种可选的实施方式中，在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，所述基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征，包括：基于所述语义分割结果，从所述目标图像中确定属于所述第一目标对象的轮廓的像素点；基于属于所述第一目标对象的 轮廓的像素点在所述目标图像中的位置，得到所述第一目标对象的轮廓线；基于所述第一目标对象的轮廓线，确定所述第一目标对象在所述目标图像中的第一二维几何特征。
该实施方式中，通过从目标图像中确定属于第一目标对象的轮廓的像素点，可以确定第一目标对象的轮廓线，降低了基于像素点确定的轮廓线边缘不整齐造成的顶点识别难度，同时在较大程度上简化了对第一目标对象的第一二维几何特征的表达，有利于相同目标对象之间的匹配。
一种可选的实施方式中，在所述第一目标对象包括线型的第三道路标志对象的情况下，所述基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征，包括：基于所述语义分割结果，拟合得到第一目标对象的中心线；基于位于所述中心线上、且属于所述第一目标对象所在图像区域的像素点在所述目标图像中的二维坐标值，确定属于所述第一目标对象所在图像区域、且位于所述中心线上的目标线段；基于所述目标线段得到所述第一目标对象的第一二维几何特征。
该实施方式中，通过确定的第一目标对象的目标线段得到第一目标对象的第一二维几何特征，可以较好的解决由于停止线以及车道实线在道路上持续地连续出现的，仅确定停止线以及车道实线的轮廓线无法对自动驾驶车辆进行准确地位姿解算的问题，以较为准确的确定第一目标对象的第一二维几何特征。
一种可选的实施方式中，还包括：基于所述目标场景中的第二目标对象在所述目标场景中的三维几何特征、所述第一目标对象的第一二维几何特征，生成所述第一目标对象和所述第二目标对象的所述对应关系。
该实施方式中，由于第二目标对象在目标场景中的三维几何特征、以及第一目标对象的第一二维几何特征均是较为简单的几何特征，因此在利用三维几何特征以及第一二维几何特征生成第一目标对象和第二目标对象的对应关系时计算量较小，处理速度较快，因此可以提高处理的效率。
一种可选的实施方式中，所述基于所述目标场景中的第二目标对象在所述目标场景中的三维几何特征、所述第一目标对象的第一二维几何特征，生成所述第一目标对象和所述第二目标对象的所述对应关系，包括：基于所述采集所述目标图像的采集设备的初始位姿信息、以及所述第二目标对象在所述目标场景中的三维几何特征，将所述第二目标对象投影至所述目标图像的图像坐标系中，得到所述第二目标对象在所述图像坐标系中的第一投影几何特征；基于所述第一目标对象在所述图像坐标系中的第一二维几何特征、以及所述第二目标对象在所述图像坐标系中的第一投影几何特征，对所述第一目标对象和所述第二目标对象进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系。
该实施方式中，通过确定的第一投影几何特征与第一二维几何特征进行匹配，一方面由于第一投影几何特征以及第一二维几何特征分别对应的几何特征均较为简单，因此在匹配时计算量较小，效率较高；另一方面，可以通过在同一个图像坐标系下较为容易地进行匹配，从而确定第一目标对象和第二目标对象的对应关系。
一种可选的实施方式中，所述基于所述目标场景中的第二目标对象在所述目标场景中的三维几何特征、所述第一目标对象的第一二维几何特征，生成所述第一目标对象和所述第二目标对象的所述对应关系，包括：基于所述目标图像与所述第二目标对象所在目标平面之间的单应矩阵，将所述目标图像中的第一目标对象投影至所述目标平面中，得到所述第一目标对象在所述目标平面中的第二投影几何特征；基于所述第一目标对象在所述目标平面中的第二投影几何特征、以及所述第二目标对象在所述目标平面中的几何特征，对所述第一目标对象和所述第二目标对象进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系；其中，所述第二目标对象在所述目标平面中的几何特征，是基于所述第二目标对象在所述场景坐标系中的三维几何特征确定的。
该实施方式中，由于利用单应矩阵可以较为准确的将第一目标对象投影至目标平面中的转换矩阵，因此将第一目标对象投影至目标平面中得到的第二投影几何特征也较为准确，使得可以利用第一目标对象在目标平面中的第二投影几何特征、以及第二目标对象在所述目标平面中的几何特征，对第一目标对象和第二目标对象进行匹配时，能够较为准确的得到第一目标对象和第二目标对象的对应关系。
一种可选的实施方式中，所述基于所述第一目标对象在所述目标平面中的第二投影几何特征、以及所述第二目标对象在所述目标平面中的几何特征，对所述第一目标对象和所述第二目标对象进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系，包括：在所述第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，利用所述第二投影几何特征中表征所述第一目标对象的顶点的特征、以及所述第二目标对象在所述目标平面中的几何特征中表征所述第二目标对象确定 的顶点的特征进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系；在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，利用所述第二投影几何特征中表征所述第一目标对象的轮廓线和/或角点的特征、以及所述第二目标对象在所述目标平面中的几何特征中表征所述第二目标对象确定的轮廓线和/或角点的特征进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系；在第一目标对象包括线型的第三道路标志对象的情况下，对所述第二投影几何特征、以及所述第二目标对象在所述目标平面中的几何特征进行最大图匹配，得到所述第一目标对象和所述第二目标对象的对应关系。
一种可选的实施方式中，所述基于位置匹配的结果，确定采集所述目标图像的采集设备的目标位姿信息，包括：基于位置匹配的结果，确定位置匹配误差；基于所述位置匹配误差以及采集所述目标图像的采集设备的初始位姿信息，确定采集所述目标图像的采集设备的目标位姿信息。
该实施方式中，利用位置匹配误差可以反应目标图像的采集设备的位姿信息的准确程度，并且，基于该匹配损失对位姿信息进行不断优化，提升目标位姿信息的精确度。
一种可选的实施方式中，所述基于所述位置匹配误差以及采集所述目标图像的采集设备的初始位姿信息，确定采集所述目标图像的采集设备的目标位姿信息，包括：检测是否满足预设的迭代停止条件；在满足所述迭代停止条件的情况下，将最后一次迭代得到的所述初始位姿信息，确定为所述目标位姿信息；在不满足所述迭代停止条件的情况下，基于所述位置匹配误差、以及最近一次迭代过程中的初始位姿信息，确定新的初始位姿信息，并返回至基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配的步骤。
该实施方式中，通过设置预设的迭代停止条件，可以使得得到的目标位姿信息可以达到较高的置信度，也即得到的目标位姿信息较为准确。
一种可选的实施方式中，所述迭代停止条件包括下述任一项：迭代次数大于预设迭代次数阈值；所述第一目标对象和所述第二目标对象的位置匹配误差小于预设的损失阈值。
一种可选的实施方式中，所述基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配，包括：在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，对所述第一目标对象的第一二维几何特征进行插值处理，得到所述第一目标对象的第二二维几何特征；其中，所述第二二维几何特征包括：多个顶点、以及多个插值点；基于所述第二二维几何特征、所述三维几何特征以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行点对点的位置匹配。
该实施方式中，通过对所述第一目标对象的第一二维几何特征进行插值处理，可以平衡不同语义之间的权重，同时也可以缓解使用较少的顶点在进行位置匹配处理时匹配不佳的问题。
一种可选的实施方式中，所述基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配，包括：基于采集所述目标图像的所述采集设备的初始位姿信息、以及所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征，将所述第二目标对象投影至所述目标图像的图像坐标系中，得到所述第二目标对象在所述图像坐标系中的第三投影几何特征；基于所述第二目标对象在所述图像坐标系中的所述第三投影几何特征、所述第一目标对象的第一二维几何特征，对具有对应关系的第一目标对象和第二目标对象进行位置匹配。
第二方面，本公开实施例提供了一种智能行驶装置的行驶控制方法，包括：获取智能行驶装置在行驶过程中采集的视频帧数据；利用第一方面或者第一方面任一种可选的实施方式中的定位方法处理所述视频帧数据，检测所述视频帧数据中的目标对象；基于检测的目标对象，控制所述智能行驶装置。
该实施方式中，由于利用本公开实施例提供的定位方法，能够更高效的确定目标位姿信息，因此该定位方法更利于部署在智能行驶装置中，提升自动驾驶控制过程中的安全性，更好的满足自动驾驶领域的需求。
第三方面，本公开实施例还提供一种定位装置，包括：第一获取模块，用于获取对目标场景进行采集得到的目标图像；第一确定模块，用于基于所述目标图像，确定所述目标图像中包括的第一 目标对象在所述目标图像中的第一二维几何特征；匹配模块，用于基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配；第二确定模块，用于基于位置匹配的结果，确定采集所述目标图像的采集设备的目标位姿信息。
第四方面，本公开实施例还提供一种智能行驶装置的行驶控制装置，包括：第二获取模块，用于获取智能行驶装置在行驶过程中采集的视频帧数据；检测模块，用于利用第一方面或者第一方面任一种可选的实施方式中的定位方法处理所述视频帧数据，检测所述视频帧数据中的目标对象；控制模块，用于基于检测的目标对象，控制所述智能行驶装置。
第五方面，本公开可选实现方式还提供一种计算机设备，处理器、存储器，所述存储器存储有所述处理器可执行的机器可读指令，所述处理器用于执行所述存储器中存储的机器可读指令，所述机器可读指令被所述处理器执行时，所述机器可读指令被所述处理器执行时执行上述第一方面或第二方面中任一种可能的实施方式中的步骤。
第六方面，本公开可选实现方式还提供一种计算机可读存储介质，该计算机可读存储介质上存储有计算机程序，该计算机程序被运行时执行上述第一方面或第二方面中任一种可能的实施方式中的步骤。
第七方面，本公开提供了一种计算机程序，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行上述第一方面或第二方面中任一种可能的实施方式中的步骤。
关于上述装置、计算机设备、及计算机可读存储介质的效果描述参见上述对应方法的说明，这里不再赘述。
为使本公开的上述目的、特征和优点能更明显易懂，下文特举实施例，并配合所附附图，作详细说明如下。
附图说明
为了更清楚地说明本公开实施例的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，此处的附图被并入说明书中并构成本说明书中的一部分，这些附图示出了符合本公开的实施例，并与说明书一起用于说明本公开的技术方案。应当理解，以下附图仅示出了本公开的某些实施例，因此不应被看作是对范围的限定，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他相关的附图。
图1示出了本公开实施例所提供的一种定位方法的流程图；
图2示出了本公开实施例所提供一种基于目标图像，确定目标图像中包括的第一目标对象在目标图像中的第一二维几何特征的具体方法的流程图；
图3示出了本公开实施例所提供的一种对目标图像进行语义分割后得到的语义分割图的示意图；
图4示出了本公开实施例所提供的一种确定第一目标对象在目标图像中的第一二维几何特征的具体方法的流程图；
图5示出了本公开实施例提供的一种基于红绿灯的语义分割结果确定红绿灯的第一二维几何特征的示意图；
图6示出了本公开实施例提供的另一种确定第一目标对象在目标图像中的第一二维几何特征的具体方法的流程图；
图7示出了本公开实施例提供的一种基于直行指示指向标识的语义分割结果确定直行指向标识的第一二维几何特征的示意图；
图8示出了本公开实施例提供的另一种确定第一目标对象在目标图像中的第一二维几何特征的具体方法的流程图；
图9示出了本公开实施例提供的一种基于车道实线的语义分割结果确定车道实线的第一二维几何特征的示意图；
图10示出了本公开实施例提供的一种确定第一目标对象和第二目标对象之间的对应关系的示意图；
图11示出了本公开实施例所提供的一种智能行驶装置的行驶控制方法的流程图；
图12示出了本公开实施例所提供的一种定位装置的示意图；
图13示出了本公开实施例所提供的一种智能行驶装置的行驶控制装置的示意图；
图14示出了本公开实施例所提供的一种计算机设备的示意图。
具体实施方式
为使本公开实施例的目的、技术方案和优点更加清楚，下面将结合本公开实施例中附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分实施例，而不是全部的实施例。通常在此处描述和示出的本公开实施例的组件可以以各种不同的配置来布置和设计。因此，以下对本公开的实施例的详细描述并非旨在限制要求保护的本公开的范围，而是仅仅表示本公开的选定实施例。基于本公开的实施例，本领域技术人员在没有做出创造性劳动的前提下所获得的所有其他实施例，都属于本公开保护的范围。
经研究发现，在基于视觉的定位方法中，通过拍摄目标场景得到的场景图像，然后从提取场景的特征点，并将提取出来的特征点、和预先基于目标场景建立的三维场景模型中的特征点进行匹配，以得到获取场景图像的图像获取设备在目标场景中的位姿信息。对于高速运动的自动驾驶车辆而言，为了保证自动驾驶车辆的安全性，需要实时、高效、精准的确定自动驾驶车辆的位姿信息，而当前基于视觉的定位方法中，基于特征点的匹配关系解算位姿信息的过程需要耗费较多的时间，效率较低，进而无法满足自动驾驶领域的需求。
基于上述研究，本公开提供了一种定位方法、行驶控制方法、装置、计算机设备及存储介质，利用了目标对象的几何特征对第一目标对象和第二目标对象进行位置匹配，并基于位置匹配的结果，得到目标图像的目标位姿信息，该方法中匹配过程简单，且处理的数据量更少，因此具有更快的处理速度，提升确定采集目标图像的采集设备的目标位姿信息的效率。
针对以上方案所存在的缺陷，均是发明人在经过实践并仔细研究后得出的结果，因此，上述问题的发现过程以及下文中本公开针对上述问题所提出的解决方案，都应该是发明人在本公开过程中对本公开做出的贡献。
应注意到：相似的标号和字母在下面的附图中表示类似项，因此，一旦某一项在一个附图中被定义，则在随后的附图中不需要对其进行进一步定义和解释。
为便于对本实施例进行理解，首先对本公开实施例所公开的一种定位方法进行详细介绍，本公开实施例所提供的定位方法的执行主体一般为具有一定计算能力的计算机设备，该计算机设备例如包括：终端设备或服务器或其它处理设备，终端设备可以为用户设备(User Equipment，UE)、移动设备、用户终端、终端、蜂窝电话、无绳电话、个人数字助理(Personal Digital Assistant，PDA)、手持设备、计算设备、车载设备、可穿戴设备等。在一些可能的实现方式中，该定位方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。
下面对本公开实施例提供的定位方法加以说明。
参见图1所示，为本公开实施例提供的定位方法的流程图，所述方法包括步骤S101～S104，其中：
S101：获取对目标场景进行采集得到的目标图像；
S102：基于目标图像，确定目标图像中包括的第一目标对象在目标图像中的第一二维几何特征；
S103：基于目标场景中的第二目标对象在目标场景对应的场景坐标系中的三维几何特征、第一目标对象的第一二维几何特征、以及第一目标对象和第二目标对象的对应关系，对第一目标对象和第二目标对象进行位置匹配；
S104：基于位置匹配的结果，确定采集目标图像的采集设备的目标位姿信息。
本公开实施例通过获取对目标场景进行采集得到的目标图像，并确定目标图像中第一目标对象在目标图像中的二维几何特征后，利用目标场景中第一目标对象和目标场景中第二目标对象的匹配关系，结合目标场景中的第二目标对象在目标场景中的三维几何特征，对第一目标对象和第二目标对象进行位置匹配，并基于匹配结果，确定目标图像的目标位姿信息，该过程利用了目标对象的几何特征对第一目标对象和第二目标对象进行位置匹配，匹配过程简单，且处理的数据量更少，因此具有更快的处理速度，提升确定采集目标图像的采集设备的目标位姿信息的效率。
下面对上述S101～S104加以详细说明。
针对上述S101，本公开实施例提供的定位方法可以应用于多种领域，例如自动驾驶领域、智能仓储领域；在应用于自动驾驶领域的情况下，目标场景例如包括自动驾驶的道路场景；在该种情况下，目标场景中的目标对象可以包括道路标志对象，例如包括以下至少一项：路牌、红绿灯、斑马线、路面指向标志、停止线、车道实线以及车道虚线，等等。
另外，在将该定位方法应用于自动驾驶领域的情况下，目标场景还可以包括停车场场景；目标场景内的目标对象例如包括：监控摄像头、感应刷卡器、车辆检测线、道路指向标志、停车线等。
在应用于智能仓储领域的情况下，目标场景例如包括仓库；在目标场景中的目标对象例如包括仓库中的货架、指示地标等。
此处，行驶车辆以及对应的目标场景可以根据实际情况确定，在此不做限定。
本公开实施例以目标场景为自动驾驶车辆行驶的道路为例，在获取目标场景的目标图像时，例如可以在自动驾驶车辆上安装图像获取设备；该图像获取设备能够实时对目标场景进行扫描，得到目标场景的目标图像。
另外，在本公开示例中，还可以在自动驾驶车辆上安装雷达、深度相机、距离传感器等其他设备，自动驾驶车辆也可以基于其他设备获取的数据进行定位，并利用基于本公开实施例提供的定位方法确定的定位结果、和利用其他设备获取的检测数据确定的定位结果结合，得到更为准确的定位结果。
针对上述S102，在获取目标图像后，即可以利用目标图像，确定目标图像中包括的第一目标对象、以及第一目标对象在目标图像中的第一二维几何特征。其中，第一目标对象例如可以包括目标图像中的路牌、红绿灯、斑马线、路面指向标志、禁止标记、停止线、车道实线以及车道虚线中至少一种；上述目标对象为目标场景中较为通用的标志性元素。由于不同的目标对象在目标图像中示出时，会呈现为不同的图形特征，因此可以根据在目标图像中的不同目标对象，确定其对应的二维几何信息。其中：
(1)：在第一目标对象包括路牌、红绿灯、斑马线、以及车道虚线中至少一种时，第一目标对象的图形特征例如包括具有矩形轮廓。此时，第一目标对象为包括具有矩形轮廓的第一道路标志对象。则第一目标对象的第一二维几何特征包括第一目标对象的顶点。例如，在第一目标对象为斑马线时，其轮廓可以利用多个连续的矩形示出，则其对应的第一二维几何特征例如可以包括其中一个矩形的顶点，或者连续示出的多个矩阵分别对应的顶点。
(2)：在第一目标对象包括路面指向标志、以及禁止标记中至少一种时，第一目标对象的图形特征例如包括具有不规则轮廓。此时，第一目标对象为包括具有不规则轮廓的第二道路标志对象。则第一目标对象的第一二维几何特征包括第一目标对象的轮廓线和/或角点。例如，第一目标对象包括路面指向标志中的右转指向标识时，其轮廓线不能由矩形和圆形等基础轮廓线直接构成，其对应的第一二维几何特征则直接对应其不规则的轮廓线。
(3)：在第一目标对象包括停止线、以及车道实线中至少一种时，第一目标对象的图形特征例如包括以线型示出。此时，第一目标对象为包括线型的第三道路标志。则第一目标对象的第一二维几何特征包括：相邻的两个第一目标对象的中心线上的目标线段。例如，在第一目标对象包括车道实线时，由于车道实线较长，因此在目标图像中可能仅能示出车道实线中的一端，或者任一端均不能示出，其对应的第一二维几何特征对应的几何特征对应属于所述第一目标对象所在图像区域、且位于所述第一目标对象的中心线上的目标线段。
此处，第一目标对象在目标图像中的第一二维几何信息，例如可以表示为第一目标对象在目标图像对应的图像坐标系中的二维坐标值。
示例性的，第一目标对象的顶点，例如可以表示为顶点在图像坐标系中的二维坐标值；轮廓线例如可以表示为轮廓线的端点在图像坐标系中的二维坐标值；目标线段可以表示为线段的端点在图像坐标系中的二维坐标值。
目标图像对应的图像坐标系，例如可以依据目标图像中任一像素点确定，例如将任一像素点作为坐标原点确定图像坐标系；具体的，可以基于图像获取设备在自动驾驶车辆中的安装位置确定；例如若图像获取设备安装在较高的位置，且视野也较高，则可以将目标图像中位置较低的像素点作为原点，确定图像坐标系。若图像获取设备安装在较低的位置，视野也较低，则可以将目标图像中位置较高的像素点作为原点确定图像坐标系。另外，也可以将图像获取设备的光轴在目标图像中的投影像素点作为原点，建立图像坐标系。具体可以根据实际情况确定，在此不再赘述。
参见图2所示，为本公开实施例提供的一种基于目标图像，确定目标图像中包括的第一目标对象在目标图像中的第一二维几何特征的具体方法，包括：
S201：对目标图像进行语义分割处理，确定目标图像中多个像素点分别对应的语义分割结果。
该实施方式中，在对目标图像进行语义分割处理时，例如可以采用下述至少一种方法：全卷积网络(Fully Convolutional Networks，FCN)、带图像卷积网络(Convolutional Neural Networks-Fully Convolutional Networks，CNN-CRF)、编码器解码器模型(Convolution-Deconvolution Networks，CDN)、以及特征金字塔模型(Feature Pyramid Network，FPN)。
在对目标图像进行语义分割处理后，即可以确定目标图像中的多个像素点分别对应的语义分割 结果。参见图3所示，为本公开实施例提供的一种对目标图像进行语义分割后得到的语义分割图的示意图。其中，图3中31所示的区域表示红绿灯、32所示的区域表示道路实线、33所示的区域表示路面指向标志、以及34所示的区域表示道路虚线。
S202：基于多个像素点分别对应的语义分割结果，以及多个像素点分别在目标图像中的位置，确定第一目标对象在目标图像中的第一二维几何特征。
此处，在根据不同像素点分别对应的语义分割结果、以及多个像素点在目标图像中的位置，确定第一目标对象在目标图像中的第一二维几何特征时，针对不同的第一目标对象，对应的确定第一目标对象在目标图像中的二维几何特征的方法也有所不同。
具体地，在确定第一目标对象在目标图像中的第一二维几何特征时：
A：针对第一目标对象包括具有矩形轮廓的第一道路标志对象的情况，例如可以采用下述图4所示的方式确定第一目标对象在目标图像中的第一二维几何特征：
S401：基于语义分割结果，从目标图像中确定属于第一目标对象的轮廓的像素点；
S402：基于属于第一目标对象的轮廓的像素点，拟合得到第一目标对象在目标图像中对应的包围框；
S403：基于包围框的顶点确定第一目标对象在目标图像中的第一二维几何特征。
示例性的，在基于语义分割结果确定属于第一目标对象的轮廓的像素点时，例如基于目标图像中各个像素点的语义分割结果，从目标图像中确定第一目标对象的所有像素点构成的区域，然后将区域边缘的像素点，确定为属于第一目标对象的轮廓的像素点。
在确定了属于第一目标对象的轮廓的像素点后，由于由属于第一目标对象的轮廓的像素点所构成的轮廓线，通常是一条呈现小幅度波动的波纹状或者锯齿状的线条。一方面，由于该轮廓线的边缘不整齐，因此在基于该轮廓线确定的顶点表征第一目标对象的几何特征时，识别难度较大；另一方面，在基于该轮廓线确定的顶点表征第一目标对象的几何特征时，数据量也较多，不利于相同目标对象之间的匹配。因此本公开实施例中。基于属于第一目标对象的轮廓的像素点，拟合得到第一目标对象在目标图像中对应的包围框，通过包围框的顶点，构成第一目标对象的第一二维几何特征，在较大程度上简化了对第一目标对象的第一二维几何特征的表达，有利于相同目标对象之间的匹配，降低识别难度。
在基于属于第一目标对象的轮廓的像素点，拟合得到第一目标对象在目标图像中对应的包围框时，例如可以基于属于第一目标对象的轮廓的像素点，确定多条直线，并对多条直线进行拟合，得到第一目标对象的包围框。
另外，由于第一目标对象在目标场景中的实际尺寸较大，较少会出现拍摄时俯仰角过大造成的变形的情况，因此在目标图像中，得到的包围框近似为矩形。
示例性的，在基于属于第一目标对象的轮廓的像素点进行拟合得到的在目标图像中对应的包围框，例如可以由属于第一目标对象的轮廓的像素点所构成的轮廓线部分重合，或者包围第一目标对象所在的区域。该包围框可以将目标图像中的第一目标对象对应的像素点包围，通过包围框的顶点能够较为准确地表征第一目标对象在目标图像中具体的第一二维几何特征，包围框的顶点在目标图像对应的图像坐标系中的坐标值，即可以表征第一目标对象在目标图像中的具体位置。
示例性的，在确定包围框后，依据矩形框中确定对顶角的二维坐标值即能够确定矩形框在目标图像中的位置，例如可以将包围框左上角顶点和右下角顶点、或者包围框左下角顶点和右下角顶点在目标图像中的二维坐标值，作为第一目标对象在目标图像中的具体位置。利用这种方式可以在保证第一目标对象的几何特征能够准确表达，同时减少后续处理过程(如将第一目标对象和第二目标对象进行位置匹配)中的数据量。
或者，例如还可以直接将矩形框的四个顶角的二维坐标值确定为第一目标对象在目标图像中的具体位置的表示。利用这种方法处理得到的第一目标对象在目标图像中的具体位置，使得目标对象的第一二维几何特征具有更高的可读性。
参见图5所示，为本公开实施例提供的一种基于红绿灯的语义分割结果确定红绿灯的第一二维几何特征的示意图；其中，图5中a表示第一目标对象包括红绿灯时，第一目标对象在目标图像中的语义分割结果的示意图；图5中b表示第一目标对象对应的包围框的示意图，51以及52表示包围框对应的两个顶点。
B：针对第一目标对象包括具有不规则轮廓的第二道路标志对象的情况，例如可以采用下述图6所示的方式确定第一目标对象在目标图像中的第一二维几何特征：
S601：基于语义分割结果，从目标图像中确定属于第一目标对象的轮廓的像素点；
S602：基于属于第一目标对象的轮廓的像素点在目标图像中的位置，得到第一目标对象的轮廓线；
S603：基于第一目标对象的轮廓线，确定第一目标对象在目标图像中的第一二维几何特征。
在具体实施中，在第一目标对象包括路面指向标识的情况下，由于第一目标对象的形状不规则，因此在确定第一目标对象的第一二维几何特征时，需要基于属于第一目标对象的轮廓的像素点确定第一目标对象的第一二维几何特征。
以路面指向标志为例，用于确定第一二维几何特征的顶点例如可以选取在包围第一目标对象的边缘中较大角度转折的转折点。由于实际的第一目标对象的轮廓线中转折点对应的位置线条起伏、转折等变化较大，因此第一目标对象的轮廓线即使呈现的小幅度波动，对顶点识别的干扰也较小，从而使得对基于语义分割结果确定的轮廓线进行识别以确定顶点时较为容易，也即可以较为方便地确定第一目标对象在目标图像中的具体位置。
示例性的，参见图7所示，为本公开实施例提供的一种基于直行指示指向标识的语义分割结果确定直行指向标识的第一二维几何特征的示意图。图7中a示出了一种直行指向标识对应的语义分割图的示意图，图7中b示出了一种第一目标对象的轮廓线的示意图。其中，轮廓线为不规则的箭头型。图7中b中71、72以及73分别表示识别得到的确定第一目标对象在目标图像中对应的轮廓线的多个顶点。
C：针对第一目标对象包括线型的第三道路标志对象的情况，例如可以采用下述图8所示的方式确定第一目标对象在目标图像中的第一二维几何特征：
S801：基于语义分割结果，拟合得到第一目标对象的中心线；
S802：基于位于中心线上、且属于第一目标对象所在图像区域的像素点在目标图像中的二维坐标值，确定属于第一目标对象所在图像区域、且位于中心线上的目标线段；
S803：基于目标线段得到第一目标对象的第一二维几何特征。
在具体实施中，在第一目标对象包括轮廓中至少一端未在所述目标图像中示出的目标对象的情况下，由于在车辆正常驾驶的过程中，停止线以及车道实线在道路上是持续地连续出现的，仅确定停止线以及车道实线的轮廓线无法对自动驾驶车辆进行准确地位姿解算，因此选用停止线以及车道实线的目标线段作为第一目标对象的第一二维几何特征。
其中，目标线段可以表示为目标线段的端点在目标图像中的二维坐标值。
示例性的，在确定第一目标对象的目标线段时，例如可以先确定第一目标对象的中心线。在基于第一目标对象的语义分割结果提取中心线时，例如可以采用下述至少一种方法：基于拓扑细化的方法、基于距离变换的方法、基于路径规划的方法、以及基于追踪的方法。示例性的，在利用基于拓扑细化的方法提取中心线时，可以先确定第一目标对象对应的语义分割图，然后利用形态学的原理对语义分割图的边界进行腐蚀消除的迭代处理，直至得到语义分割图中第一目标对象对应的中心线。由于不同的提取中心线的方法适用场景以及对图像的质量要求不同，因此具体提取中心线方法的选取和具体的执行过程可以根据实际情况确定，在此不再赘述。
在确定第一目标对象的中心线后，还可以在中心线上确定两个点，作为目标线段的端点。
具体地，在确定中心线上目标线段的端点时，例如可以采用逐点搜索的方法，逐一确定中心线上的点是否在第一目标对象对应的图像区域中有对应的像素点；或者在第一目标对象对应的图像区域中确定位于中心线上的像素点。此时，即可确定属于第一目标对象所在图像区域、且位于中心线上的所有像素点，并将距离最远的两个像素点作为目标线段的端点。然后，将距离最远的两个像素点确定的目标线段，作为第一目标对象的第一二维几何特征。
参见图9所示，为本公开实施例提供的一种基于车道实线的语义分割结果确定车道实线的第一二维几何特征的示意图。其中，图9中a表示车道实线的语义分割图，图9中b表示车道实线对应的目标线段。
此时，在完成上述A、B以及C中至少一种的情况下，即可确定第一目标对象的第一二维几何特征。在具体实施中，在第一目标对象包括第一道路标志对象和第二道路标志对象中至少一项时，第一目标对象的第一二维几何特征例如可以表示为p j；其中，j表示多个第一目标对象中的第j个第一目标对象。在第一目标对象包括第三道路标志对象中至少一种时，第一目标对象的第一二维几何特征例如可以表示为l i；其中，i表示多个第一目标对象中的第i个第一目标对象。
针对上述S103，第二目标对象与第一目标对象对应，例如可以包括路牌、红绿灯、斑马线、车道实线、路面指向标志、禁止标记、停止线、以及车道虚线中至少一项。第二目标对象所在的目标 场景对应的场景坐标系，例如可以包括预先建立的场景坐标系。其中，场景坐标系是针对目标场景建立的三维坐标系。具体地，可以直接选用世界坐标系作为场景坐标系，或者以目标场景中任一位置点作为原点，建立场景坐标系。具体的场景坐标系可以根据实际情况确定，在此不再赘述。
在确定了目标场景的场景坐标系的情况下，可以确定目标场景中的第二目标对象在目标场景对应的场景坐标系中的三维几何特征。其中，第二目标对象的三维几何特征例如可以是预先确定的，示例性的，例如可以利用同步定位与建图(Simultaneous Localization and Mapping，SLAM)建模、运动恢复结构(Structure-From-Motion，SFM)建模中任一种方法确定目标场景中的第二目标对象的三维几何特征。具体确定第二目标对象的三维几何特征的方法可以根据实际情况确定，在此不再赘述。
在具体实施中，在第二目标对象包括第一道路标志对象和第二道路标志对象中至少一项的情况，第二目标对象的三维几何特征例如包括第二目标对象中的各个顶点。另外，对于第二目标对象包括具有不规则轮廓的第二道路标志对象的情况，对应的三维几何特征例如还可以包括第二目标对象的角点。
其中，可以利用各个顶点在三维坐标系中的三维坐标值表示第二目标对象的三维几何特征。
第二目标对象包括第三道路标志对象的情况，第二目标对象的三维几何特征例如包括位于第二目标对象的中线，且属于第二目标对象的线段。其中，可以利用线段的端点在三维坐标系中的三维坐标值表示第二目标对象的三维几何特征。
在具体实施中，在第二目标对象包括轮廓利用至少一个矩形示出的目标对象或利用不规则图形示出的目标对象中至少一项时，第二目标对象的三维几何特征例如可以表示为P j；其中，j表示多个第二目标对象中的第j个第二目标对象。在第二目标对象包括轮廓中至少一端未在所述目标图像中示出的目标对象中至少一种时，第二目标对象的三维几何特征例如可以表示为L i；其中，i表示多个第二目标对象中的第i个第二目标对象。
此时，还可以基于目标场景中的第二目标对象在目标场景中的三维几何特征、第一目标对象的第一二维几何特征，生成第一目标对象和第二目标对象的对应关系。
本公开实施例还提供一种生成第一目标对象和第二目标对象的对应关系的具体方法，包括：基于目标场景中的第二目标对象在目标场景中的三维几何特征、第一目标对象的第一二维几何特征，生成第一目标对象和第二目标对象的对应关系。
示例性的，第二目标对象包括：轮廓利用至少一个矩形示出的目标对象或利用不规则图形示出的目标对象中至少一项的情况，可以采用下述方式生成的第一目标对象和第二目标对象之间的对应关系：
基于采集目标图像的采集设备的初始位姿信息、以及第二目标对象在目标场景中的三维几何特征，将第二目标对象投影至目标图像的图像坐标系中，得到第二目标对象在图像坐标系中的第一投影几何特征；
基于第一目标对象在图像坐标系中的第一二维几何特征、以及第二目标对象在图像坐标系中的第一投影几何特征，对第一目标对象和第二目标对象进行匹配，得到第一目标对象和第二目标对象的对应关系。
在具体实施中，在获取采集目标图像的采集设备的初始位姿信息时，由于自动驾驶车辆在目标场景中的道路上正常行驶时，位姿通常不会发生较大幅度的变化，因此例如可以预先获取先前确定的自动驾驶车辆在道路上的位姿信息，作为初始位姿信息；或者，由于目标场景中道路的信息可以较容易的得到，例如采用全球定位系统(Global Positioning System，GPS)确定道路的位置信息，然后基于目标场景中道路的位置信息估算得到自动驾驶车辆的初始位姿信息。其中，初始位姿信息例如可以表示为T 0。具体地确定目标图像的初始位姿的信息的方法可以按照实际情况确定，在此不再赘述。
此时，由于目标图像例如可以是基于自动驾驶车辆上安装的图像采集设备得到的，因此采集目标图像的采集设备的位姿信息与自动驾驶车辆对应的位姿信息相关联。在忽略图像采集设备与自动驾驶车辆的相对位姿关系的情况下，或者可以通过其他位姿解算方式，例如基于图像采集设备与自动驾驶车辆的相对位姿关系能够确定自动驾驶车辆的位姿信息的情况下，基于采集目标图像的采集设备位姿信息即可确定自动驾驶车辆的位姿信息。
在确定采集目标图像的采集设备的初始位姿信息、以及第二目标对象在目标场景中的三维几何特征后，通过将第二目标对象投影至目标图像的图像坐标系中，可以得到第二目标对象在图像坐标 系中的第一投影几何特征。
在具体实施中，将第二目标对象投影至目标图像的图像坐标系中时，也即将第二目标对象的三维几何特征投影至目标图像对应的图像坐标系中，例如可以采用下述方式：将第二目标对象的三维几何特征由场景坐标系转换至世界坐标系，完成由场景空间坐标到世界空间坐标的模型空间转换；然后基于采集目标图像的采集设备的初始位姿信息，将转换至世界坐标系下的三维几何特征由世界坐标系转换至图像坐标系，完成由世界空间坐标到相机空间坐标的观察空间转化。此时，即可以得到在图像坐标系中第二目标对象的第一投影几何特征。具体确定第二目标对象的第一投影几何特征的方法可以根据实际情况确定，在此不再赘述。
此时，由于目标图像中的第一目标对象与目标场景中的第二目标对象之间存在对应关系，因此得到的第二目标对象在图像坐标系中的第一投影几何特征与第一目标对象在图像坐标系中的第一二维几何特征之间也存在对应关系，从而可以基于第一目标对象在图像坐标系中的第一二维几何特征、以及第二目标对象在图像坐标系中的第一投影几何特征，对第一目标对象和第二目标对象进行匹配，以得到第一目标对象和第二目标对象的对应关系。
其中，在对第一目标对象和第二目标对象进行匹配时，例如可以采用最近邻匹配(k-Nearest Neighbor，KNN)或者其他匹配方法确定第一目标对象和第二目标对象的对应关系，具体的匹配过程在此不再赘述。
此时，即可确定在第二目标对象包括轮廓利用至少一个矩形示出的目标对象或利用不规则图形示出的目标对象中至少一项时，第一目标对象和第二目标对象的对应关系。
在第二目标对象包括轮廓利用至少一个矩形示出的目标对象或利用不规则图形示出的目标对象中至少一种时，生成第一目标对象和第二目标对象的对应关系的具体方法，包括：
基于目标图像以及第二目标对象所在目标平面之间的单应矩阵，将目标图像中的第一目标对象投影至目标平面中，得到第一目标对象在目标平面中的第二投影几何特征；基于第一目标对象在目标平面中的第二投影几何特征、以及第二目标对象在目标平面中的几何特征，对第一目标对象和第二目标对象进行匹配，得到第一目标对象和第二目标对象的对应关系。其中，第二目标对象在目标平面中的几何特征，是基于第二目标对象在场景坐标系中的三维几何特征确定的。
其中，目标图像以及第二目标对象所在目标平面之间的单应矩阵(homography)用于将目标图像中的第一目标对象投影至目标平面中。
示例性的，在获取目标图像以及第二目标对象所在目标平面之间的单应矩阵时，例如可以获取第二目标对象在目标平面上的任一像素点的坐标值，其中，第二目标对象在目标平面中的几何特征，是基于第二目标对象在场景坐标系中的三维几何特征确定的，例如表示为O 1＝(x 1,y 1,z 1)；以及获取第二目标对象包括的至少两个像素点在目标图像上对应位置的坐标系O 2＝(x 2,y 2,z 2)，然后例如可以根据下述公式(1)确定单应矩阵，例如可以用H表示：
O 2＝HO 1        (1)
其中，由于在坐标值O 1中的z 1、以及O 2中的z 2在平面中为0，因此在计算单应矩阵时，可以设置为z 1＝1,z 2＝1进行计算。获取单应矩阵H的具体过程在此不再赘述。
在确定单应矩阵H后，例如可以将第一目标对象投影至目标平面中，以获取第一目标对象在目标平面中的第二投影几何特征。
在确定第二投影几何特征后，即可以基于第一目标对象在图像坐标系中的第一二维几何特征、以及第一目标对象在目标平面中的第二投影几何特征，对第一目标对象和第二目标对象进行匹配，得到第一目标对象和第二目标对象的对应关系。
在具体实施中，在第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，利用第二投影几何特征中表征第一目标对象的顶点的特征、以及第二目标对象在目标平面中的几何特征中表征第二目标对象确定的顶点的特征进行匹配，得到第一目标对象和第二目标对象的对应关系。
此处，由于第一目标对象包括具有矩形轮廓的第一道路标志，通过第二投影几何特征中表征第一目标对象的顶点的特征、以及第二目标对象在目标平面中的几何特征中表征第二目标对象确定的顶点的特征进行匹配，即可以较为简单的确定第一目标对象和第二目标对象的对应关系。通过采用这样以几何特征中的点进行匹配的方式，对于第一目标对象包括具有矩形轮廓的第一道路标志对象，在运算量少的同时，准确性也较高。
类似的，在第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，利用第二投影几 何特征中表征第一目标对象的轮廓线和/或角点的特征、以及第二目标对象在目标平面中的几何特征中表征第二目标对象确定的轮廓线和/或角点的特征进行匹配，得到第一目标对象和第二目标对象的对应关系。
此处，由于第一目标对象包括具有不规则轮廓的第二道路标志对象，对应的将第二投影几何特征中表征第一目标对象的轮廓线和/或角点的特征、以及第二目标对象在目标平面中的几何特征中表征第二目标对象确定的轮廓线和/或角点的特征进行匹配的方式，可以与第一道路标志对应的匹配方式相似，采用轮廓线中的轮廓线顶点和/或直接采用确定的角点进行匹配。这样，对于较之第一道路标志对象而言图形表达更复杂的第二道路标志对象，可以相应的提高其匹配时的准确性。
在第一目标对象包括线型的第三道路标志对象的情况下，对第二投影几何特征、以及第二目标对象在目标平面中的几何特征进行最大图匹配，得到第一目标对象和第二目标对象的对应关系。
具体地，针对线型的第三道路标志，采用最大图匹配算法对第一目标对象和第二目标对象进行匹配的过程中，可以随机地以任意匹配方式构建多个候选匹配对，每个候选匹配对包含一个第一目标对象和一个第二目标对象，且每个第一目标对象仅包含于一个候选匹配对，每个第二对象仅包含于一个候选匹配对。可以计算各个候选匹配对中第一目标对象和第二目标对象的直线距离，确定各个候选匹配对对应的上述直线距离的平均值。在匹配过程中，去除上述平均值过大的匹配方式，由此获得第一目标对象和第二目标对象的匹配关系。该方式在待匹配的第一目标对象的数量与第二目标对象的数量不相同的情况下也可以完成匹配过程，获得数量较少的第一目标对象与数量较多的第二目标对象的匹配结果，由此通过最大图匹配实现了对目标图像中漏检的线型的第三道路标志的自动识别，能够获得比较准确的匹配结果。
此处，由于第一目标对象包括线型的第三道路标志对象，由于其在投影后不能示出其包括的所有点的信息，尤其是表征其实际位置的所有点。因此上述利用点、轮廓线、角点中至少一种对第一道路标志对象和/或第二道路标志对象进行匹配的方式并不适用。而通过最大图匹配的方式，可以对第二投影几何特征、以及第二目标对象在目标平面中的几何特征直接进行匹配，而不需要重新对第三道路标志对象进行采集，以保证定位的效率。
示例性的，在对第一目标对象和第二目标对象进行匹配时，由于目标图像的初始位姿信息并不准确，且道路中的第二目标对象具有连续出现的特点，因此利用最近邻匹配得到的对应关系并不准确，从而在匹配时容易出现漏检线条，使得确定的第一目标对象和第二目标对象的对应关系不准确。因此，例如可以选用最优匹配算法(Kuhn-Munkras，KM)对第一目标对象和第二目标对象进行匹配，通过将第一目标对象中任一两个目标线段之间的距离的倒数作为两个目标线段对应的第一目标对象的权重值，可以去除匹配候选中对目标线段距离平均值过大的匹配，从而获取第一目标对象和第二目标对象之间的对应关系。
参见图10所示，为本公开实施例提供的一种确定第一目标对象和第二目标对象之间的对应关系的示意图。其中，11表示第一目标对象，12表示第二目标对象，13表示通过最大图匹配算法对线条进行匹配后发现的不存在与之相匹配的第二目标对象的第一目标对象，即13表示在目标图像中漏检的线条，14表示一组对应的第一目标对象和第二目标对象。其中，在存在漏检线条的情况下，例如可以不对漏检线条进行处理。
在确定了第一目标对象和第二目标对象之间的对应关系后，即可以基于目标场景中的第二目标对象在目标场景对应的场景坐标系中的三维几何特征、第一目标对象的第一二维几何特征，对第一目标对象和第二目标对象进行位置匹配。
在对第一目标对象和第二目标对象进行位置匹配时，由于已经确定了第一目标对象和第二目标对象的对应关系，通过第二目标对象在场景坐标系中的三维几何特征与第一目标对象的第一二维几何特征进行位置匹配，可以确定由于初始位姿信息与实际位姿信息的偏差造成的匹配损失，从而基于确定的匹配损失确定采集目标图像的采集设备的目标位姿信息。
基于采集目标图像的采集设备的初始位姿信息、以及目标场景中的第二目标对象在目标场景对应的场景坐标系中的三维几何特征，将第二目标对象投影至图像坐标系中，得到第二目标对象在图像坐标系中的第三投影几何特征。基于第二目标对象在图像坐标系中的第三投影几何特征、第一目标对象的第一二维几何特征、以及第一目标对象和第二目标对象的对应关系，对具有对应关系的第一目标对象和第二目标对象进行位置匹配。
在具体实施中，确定第二目标对象在图像坐标系中的第三投影几何特征的方法与上述确定第一投影几何特征的方式相似，在此不再赘述。在第一目标对象包括轮廓中至少一端未在所述目标图像中示出的目标对象中中至少一种的情况下，确定的第三投影几何特征，例如可以表示为π(L i,T 0)； 其中，π为投影函数，用于将第二目标对象的三维几何特征L i根据初始位姿信息T 0投影至图像坐标系中。在第一目标对象包括轮廓利用至少一个矩形示出的目标对象或利用不规则图形示出的目标对象中至少一项的情况下，确定的第三投影几何特征，例如可以表示为π(P j,T 0)；其中，投影函数π用于将第二目标对象的三维几何特征P j根据初始位姿信息T 0投影至图像坐标系中。
此时，例如还可以确定对应于第一目标对象中顶点或者目标线段，第二目标对象中的投影顶点或者投影目标线段。具体确定投影顶点或者投影目标线段的方法，与上述确定第一目标对象中顶点或者目标线段的方法相似，在此不再赘述。
在第一目标对象包括轮廓中至少一端未在所述目标图像中示出的目标对象中中至少一种的情况下，根据第一目标对象和第二目标对象的对应关系，可以将第一目标对象的第一二维几何特征，与对应的第二目标对象在图像坐标系中的第三投影几何特征进行位置匹配，从而确定第一目标对象与第二目标对象的对应关系。
此时，由于第一目标对象的第一二维几何特征是通过第一目标对象的目标线段的端点坐标值确定的，因此在确定对应关系时，只需要对端点进行位置匹配，运算量更少，从而使得在位置匹配时效率更高。
在第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，为本公开实施例提供的一种对第一目标对象和第二目标对象进行位置匹配的具体方法，包括：对第一目标对象的第一二维几何特征进行插值处理，得到第一目标对象的第二二维几何特征；其中，第二二维几何特征包括：多个顶点在目标图像中的坐标值、以及多个插值点在目标图像中的坐标值；基于第二二维几何特征、三维几何特征以及第一目标对象和第二目标对象的对应关系，对第一目标对象和第二目标对象进行点对点的位置匹配。
在具体实施中，在第一目标对象包括轮廓利用至少一个矩形示出的目标对象至少一种的情况下，由于第一目标对象的第一二维几何特征例如可以是基于在第一目标对象上的两个或者四个顶点的坐标值得到的，顶点可能较少，通过对各个顶点进行插值处理，确定的多个插值点可以形成第一目标对象的稀疏轮廓线，使得在图像坐标系上两个坐标轴方向上的多个顶点数目差距较小，从而平衡不同语义之间的权重，同时也可以缓解使用较少的顶点在进行位置匹配处理时匹配不佳的问题。
在对各个顶点进行插值处理时，例如可以采用下述至少一种方法：泰勒插值(Taylor Interpolation)、拉格朗日插值(Lagrange Interpolation)、牛顿插值(Newton Interpolation)、以及艾尔米特插值(Hermite Interpolation)。具体的插值方法可以根据实际情况选取，在此不再赘述。
针对上述S104，在基于位置匹配结果，确定采集目标图像的采集设备的目标位姿信息时，例如可以基于位置匹配的结果，确定位置匹配误差，并基于位置匹配误差以及采集目标图像的采集设备的初始位姿信息，确定采集目标图像的采集设备的目标位姿信息。
在具体实施中，在确定了第三投影几何特征π(L i,T 0)以及π(P j,T 0)的情况下，可以确定对应的位置匹配误差。其中，第三投影几何特征π(L i,T 0)对应的位置匹配误差例如可以表示为D l(π(L i,T 0),l i)，D l表示投影目标线段的端点到目标线段所在中心线的距离的残差项。第三投影几何特征π(P j,T 0)对应的位置匹配误差例如可以表示为D p(π(P j,T 0),p i)，D p表示顶点与投影顶点之间的重投影误差。
此时，由于可能存在多个第一目标对象，确定位置匹配误差时例如可以采用下述公式(2)：
Figure PCTCN2021098964-appb-000001
其中，Q表示第一目标对象中包括的轮廓利用至少一个矩形示出的目标对象或利用不规则图形示出的目标对象的总数量；P表示第一目标对象中包括的轮廓中至少一端未在所述目标图像中示出的目标对象中的总数量；error表示确定的匹配损失。
此时，匹配损失error越小，表征采集目标图像的采集设备的位姿信息较为准确；匹配损失error越大，表征采集目标图像的采集设备的位姿信息与实际的位姿信息差异较大，需要进一步确定目标图像的采集设备的位姿信息，以使匹配损失error减小。
在确定匹配损失的情况下，基于匹配损失确定采集目标图像的采集设备的目标位姿信息时，例如可以采用下述方法：检测是否满足预设的迭代停止条件；在满足迭代停止条件的情况下，将最后一次迭代得到的初始位姿信息，确定为目标位姿信息；在不满足迭代停止条件的情况下，基于位置 匹配误差、以及最近一次迭代过程中的初始位姿信息，确定新的初始位姿信息，并返回至基于目标场景中的第二目标对象在目标场景对应的场景坐标系中的三维几何特征、第一目标对象的第一二维几何特征、以及第一目标对象和第二目标对象的对应关系，对第一目标对象和第二目标对象进行位置匹配的步骤。
其中，迭代停止条件包括下述至少一种：迭代次数大于预设迭代次数阈值；第一目标对象和第二目标对象的位置匹配误差小于预设的损失阈值。在选取迭代停止条件为迭代次数大于预设迭代次数阈值的情况下，例如可以基于经验确定预设迭代次数阈值，例如6次或者8次，以使得在迭代足够多次数后的匹配损失较小。在选取迭代停止条件为第一目标对象和第二目标对象的位置匹配误差小于预设的损失阈值的情况下，可以通过设置一个较小的损失阈值，使得得到的目标位姿信息置信度更高。具体地迭代停止条件的选取可以根据实际情况确定，在此不再赘述。
在不满足迭代停止条件的情况下，基于此时确定的位置匹配误差，确定迭代的方向为将位置匹配误差减少的方向，并将最近一次迭代过程中的初始位姿信息确定为新的初始位姿信息，然后返回对第一目标对象和第二目标对象进行位置匹配的步骤，重新根据新的初始位姿信息确定匹配损失，直至位置匹配误差满足迭代停止条件。在满足迭代停止条件的情况下，可以将此时的初始位姿信息确定为目标位姿信息，例如可以表示为T aim。
此时，即可以确定目标位姿信息T aim，也即目标图像的目标位姿信息T aim。
基于同一发明构思，本公开实施例中还提供了一种智能行驶装置的行驶控制方法。
参见图11所示，为本公开实施例提供的一种智能行驶装置的行驶控制方法的流程图，智能行驶装置的行驶控制方法包括步骤S1101～S1103，其中：
S1101：获取智能行驶装置在行驶过程中采集的视频帧数据；
S1102：利用本公开实施例提供的定位方法得到目标检测神经网络，检测视频帧数据中的目标对象；
S1103：基于检测的目标对象，控制智能行驶装置。
在具体实施中，行驶装置例如但不限于下述任一种：自动驾驶车辆、装有高级驾驶辅助系统(Advanced Driving Assistance System，ADAS)的车辆、或者机器人等。
控制行驶装置，例如包括控制行驶装置加速、减速、转向、制动等，或者可以播放语音提示信息，以提示驾驶员控制行驶装置加速、减速、转向、制动等。
此处，由于利用本公开实施例提供的定位方法，能够更高效的确定目标位姿信息，因此该定位方法更利于部署在智能行驶装置中，提升自动驾驶控制过程中的安全性，更好的满足自动驾驶领域的需求。
本领域技术人员可以理解，在具体实施方式的上述方法中，各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定，各步骤的具体执行顺序应当以其功能和可能的内在逻辑确定。
基于同一发明构思，本公开实施例中还提供了与定位方法对应的定位装置，由于本公开实施例中的装置解决问题的原理与本公开实施例上述定位方法相似，因此装置的实施可以参见方法的实施，重复之处不再赘述。
参照图12所示，为本公开实施例提供的一种定位装置的示意图，所述装置包括：第一获取模块121、第一确定模块122、匹配模块123、以及第二确定模块124；其中，
第一获取模块121，用于获取对目标场景进行采集得到的目标图像；第一确定模块122，用于基于所述目标图像，确定所述目标图像中包括的第一目标对象在所述目标图像中的第一二维几何特征；匹配模块123，用于基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配；第二确定模块124，用于基于位置匹配的结果，确定采集所述目标图像的采集设备的目标位姿信息。
一种可选的实施方式中，在所述第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，所述第一目标对象的第一二维几何特征，包括：所述第一目标对象的顶点；在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，所述第一目标对象的第一二维几何特征包括：所述第一目标对象的轮廓线和/或角点；在所述第一目标对象包括线型的第三道路标志对象的情况下，所述第一目标对象的第一二维几何特征，包括：属于所述第一目标对象所在图像区域、且位于所述第一目标对象的中心线上的目标线段。
一种可选的实施方式中，所述第一确定模块122在基于所述目标图像，确定所述目标图像中包 括的第一目标对象在所述目标图像中的第一二维几何特征时，用于：对所述目标图像进行语义分割处理，确定所述目标图像中多个像素点分别对应的语义分割结果；基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征。
一种可选的实施方式中，在所述第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，所述第一确定模块122在基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征时，用于：基于所述语义分割结果，从所述目标图像中确定属于所述第一目标对象的轮廓的像素点；基于属于所述第一目标对象的轮廓的像素点，拟合得到所述第一目标对象在所述目标图像中对应的包围框；基于所述包围框的顶点确定所述第一目标对象在所述目标图像中的第一二维几何特征。
一种可选的实施方式中，在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，所述第一确定模块122在基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征时，用于：基于所述语义分割结果，从所述目标图像中确定属于所述第一目标对象的轮廓的像素点；基于属于所述第一目标对象的轮廓的像素点在所述目标图像中的位置，得到所述第一目标对象的轮廓线；基于所述第一目标对象的轮廓线，确定所述第一目标对象在所述目标图像中的第一二维几何特征。
一种可选的实施方式中，在所述第一目标对象包括线型的第三道路标志对象的情况下，所述第一确定模块122在基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征时，用于：基于所述语义分割结果，拟合得到第一目标对象的中心线；基于位于所述中心线上、且属于所述第一目标对象所在图像区域的像素点在所述目标图像中的二维坐标值，确定属于所述第一目标对象所在图像区域、且位于所述中心线上的目标线段；基于所述目标线段得到所述第一目标对象的第一二维几何特征。
一种可选的实施方式中，还包括生成模块125，用于：基于所述目标场景中的第二目标对象在所述目标场景中的三维几何特征、所述第一目标对象的第一二维几何特征，生成所述第一目标对象和所述第二目标对象的所述对应关系。
一种可选的实施方式中，所述生成模块125在基于所述目标场景中的第二目标对象在所述目标场景中的三维几何特征、所述第一目标对象的第一二维几何特征，生成所述第一目标对象和所述第二目标对象的所述对应关系时，用于：基于所述采集所述目标图像的采集设备的初始位姿信息、以及所述第二目标对象在所述目标场景中的三维几何特征，将所述第二目标对象投影至所述目标图像的图像坐标系中，得到所述第二目标对象在所述图像坐标系中的第一投影几何特征；基于所述第一目标对象在所述图像坐标系中的第一二维几何特征、以及所述第二目标对象在所述图像坐标系中的第一投影几何特征，对所述第一目标对象和所述第二目标对象进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系。
一种可选的实施方式中，所述生成模块125在基于所述目标场景中的第二目标对象在所述目标场景中的三维几何特征、所述第一目标对象的第一二维几何特征，生成所述第一目标对象和所述第二目标对象的所述对应关系时，用于：基于所述目标图像与所述第二目标对象所在目标平面之间的单应矩阵，将所述目标图像中的第一目标对象投影至所述目标平面中，得到所述第一目标对象在所述目标平面中的第二投影几何特征；基于所述第一目标对象在所述目标平面中的第二投影几何特征、以及所述第二目标对象在所述目标平面中的几何特征，对所述第一目标对象和所述第二目标对象进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系；其中，所述第二目标对象在所述目标平面中的几何特征，是基于所述第二目标对象在所述场景坐标系中的三维几何特征确定的。
一种可选的实施方式中，所述生成模块125在基于所述第一目标对象在所述目标平面中的第二投影几何特征、以及所述第二目标对象在所述目标平面中的几何特征，对所述第一目标对象和所述第二目标对象进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系时，用于：在所述第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，利用所述第二投影几何特征中表征所述第一目标对象的顶点的特征、以及所述第二目标对象在所述目标平面中的几何特征中表征所述第二目标对象确定的顶点的特征进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系；在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，利用所述第二投影几何特征中表征所述第一目标对象的轮廓线和/或角点的特征、以及所述第二目标对象在所述目标平面中的几何特征中表征所述第二目标对象确定的轮廓线和/或角点的特征进行匹配，得到所述第一目标对象 和所述第二目标对象的对应关系；在第一目标对象包括线型的第三道路标志对象的情况下，对所述第二投影几何特征、以及所述第二目标对象在所述目标平面中的几何特征进行最大图匹配，得到所述第一目标对象和所述第二目标对象的对应关系。
一种可选的实施方式中，所述第二确定模块124在基于位置匹配的结果，确定采集所述目标图像的采集设备的目标位姿信息时，用于：基于位置匹配的结果，确定位置匹配误差；基于所述位置匹配误差以及采集所述目标图像的采集设备的初始位姿信息，确定采集所述目标图像的采集设备的目标位姿信息。
一种可选的实施方式中，所述第二确定模块124在基于所述位置匹配误差以及采集所述目标图像的采集设备的初始位姿信息，确定采集所述目标图像的采集设备的目标位姿信息时，用于：检测是否满足预设的迭代停止条件；在满足所述迭代停止条件的情况下，将最后一次迭代得到的所述初始位姿信息，确定为所述目标位姿信息；在不满足所述迭代停止条件的情况下，基于所述位置匹配误差、以及最近一次迭代过程中的初始位姿信息，确定新的初始位姿信息，并返回至基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配的步骤。
一种可选的实施方式中，所述迭代停止条件包括下述任一项：迭代次数大于预设迭代次数阈值；所述第一目标对象和所述第二目标对象的位置匹配误差小于预设的损失阈值。
一种可选的实施方式中，所述匹配模块123在基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配时，用于：在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，对所述第一目标对象的第一二维几何特征进行插值处理，得到所述第一目标对象的第二二维几何特征；其中，所述第二二维几何特征包括：多个顶点、以及多个插值点；基于所述第二二维几何特征、所述三维几何特征以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行点对点的位置匹配。
一种可选的实施方式中，所述匹配模块123在基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配时，用于：基于采集所述目标图像的采集设备的初始位姿信息、以及所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征，将所述第二目标对象投影至所述目标图像的图像坐标系中，得到所述第二目标对象在所述图像坐标系中的第三投影几何特征；基于所述第二目标对象在所述图像坐标系中的所述第三投影几何特征、所述第一目标对象的第一二维几何特征，对具有对应关系的第一目标对象和第二目标对象进行位置匹配。
关于定位装置中的各模块的处理流程、以及各模块之间的交互流程的描述可以参照上述定位方法实施例中的相关说明，这里不再详述。
基于同一发明构思，本公开实施例中还提供了与智能行驶装置的行驶控制对应的智能行驶装置的行驶控制装置，由于本公开实施例中的装置解决问题的原理与本公开实施例上述定位方法相似，因此装置的实施可以参见方法的实施，重复之处不再赘述。
参照图13所示，为本公开实施例提供的一种智能行驶装置的行驶控制装置的示意图，所述装置包括：第二获取模块131、检测模块132、以及控制模块133；其中，
第二获取模块131，用于获取智能行驶装置在行驶过程中采集的视频帧数据；
检测模块132，用于利用基于本公开实施例提供的任一种定位方法处理所述视频帧数据，检测所述视频帧数据中的目标对象；
控制模块133，用于基于检测的目标对象，控制所述智能行驶装置。
关于智能行驶装置的行驶控制装置中的各模块的处理流程、以及各模块之间的交互流程的描述可以参照上述智能行驶装置的行驶控制方法实施例中的相关说明，这里不再详述。
本公开实施例还提供了一种计算机设备，如图14所示，为本公开实施例提供的计算机设备结构示意图，包括：
处理器141和存储器142；所述存储器142存储有处理器141可执行的机器可读指令，处理器141用于执行存储器142中存储的机器可读指令，所述机器可读指令被处理器141执行时，处理器141执行下述步骤：
获取对目标场景进行采集得到的目标图像；基于所述目标图像，确定所述目标图像中包括的第一目标对象在所述目标图像中的第一二维几何特征；基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配；基于位置匹配的结果，确定采集所述目标图像的采集设备的目标位姿信息。
或者，处理器141执行下述步骤：
获取智能行驶装置在行驶过程中采集的视频帧数据；利用基于本公开实施例提供的任一种定位方法处理所述视频帧数据，检测所述视频帧数据中的目标对象；基于检测的目标对象，控制所述智能行驶装置。
上述存储器142包括内存1421和外部存储器1422；这里的内存1421也称内存储器，用于暂时存放处理器141中的运算数据，以及与硬盘等外部存储器1422交换的数据，处理器141通过内存1421与外部存储器1422进行数据交换。
上述指令的具体执行过程可以参考本公开实施例中所述的定位方法、或者智能行驶装置的行驶控制方法的步骤，此处不再赘述。
本公开实施例还提供一种计算机可读存储介质，该计算机可读存储介质上存储有计算机程序，该计算机程序被处理器运行时执行上述方法实施例中所述的定位方法、或者智能行驶装置的行驶控制方法的步骤。其中，该存储介质可以是易失性或非易失的计算机可读取存储介质。
本公开实施例还提供一种计算机程序产品，该计算机程序产品承载有程序代码，所述程序代码包括的指令可用于执行上述方法实施例中所述的定位方法、或者智能行驶装置的行驶控制方法的步骤，具体可参见上述方法实施例，在此不再赘述。
其中，上述计算机程序产品可以具体通过硬件、软件或其结合的方式实现。在一个可选实施例中，所述计算机程序产品具体体现为计算机存储介质，在另一个可选实施例中，计算机程序产品具体体现为软件产品，例如软件开发包(Software Development Kit，SDK)等等。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统和装置的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。在本公开所提供的几个实施例中，应该理解到，所揭露的系统、装置和方法，可以通过其它的方式实现。以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，又例如，多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些通信接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本公开各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个处理器可执行的非易失的计算机可读取存储介质中。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本公开各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(Read-Only Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
最后应说明的是：以上所述实施例，仅为本公开的具体实施方式，用以说明本公开的技术方案，而非对其限制，本公开的保护范围并不局限于此，尽管参照前述实施例对本公开进行了详细的说明，本领域的普通技术人员应当理解：任何熟悉本技术领域的技术人员在本公开揭露的技术范围内，其依然可以对前述实施例所记载的技术方案进行修改或可轻易想到变化，或者对其中部分技术特征进行等同替换；而这些修改、变化或者替换，并不使相应技术方案的本质脱离本公开实施例技术方案的精神和范围，都应涵盖在本公开的保护范围之内。因此，本公开的保护范围应所述以权利要求的保护范围为准。

Claims (21)

  1. 一种定位方法，其特征在于，包括：
    获取对目标场景进行采集得到的目标图像；
    基于所述目标图像，确定所述目标图像中包括的第一目标对象在所述目标图像中的第一二维几何特征；
    基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配；
    基于位置匹配的结果，确定采集所述目标图像的采集设备的目标位姿信息。
  2. 根据权利要求1所述的定位方法，其特征在于，在所述第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，所述第一目标对象的第一二维几何特征，包括：所述第一目标对象的顶点；
    在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，所述第一目标对象的第一二维几何特征包括：所述第一目标对象的轮廓线和/或角点；
    在所述第一目标对象包括线型的第三道路标志对象的情况下，所述第一目标对象的第一二维几何特征，包括：属于所述第一目标对象所在图像区域、且位于所述第一目标对象的中心线上的目标线段。
  3. 根据权利要求2所述的定位方法，其特征在于，所述基于所述目标图像，确定所述目标图像中包括的第一目标对象在所述目标图像中的第一二维几何特征，包括：
    对所述目标图像进行语义分割处理，确定所述目标图像中多个像素点分别对应的语义分割结果；
    基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征。
  4. 根据权利要求3所述的定位方法，其特征在于，在所述第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，所述基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征，包括：
    基于所述语义分割结果，从所述目标图像中确定属于所述第一目标对象的轮廓的像素点；
    基于属于所述第一目标对象的轮廓的像素点，拟合得到所述第一目标对象在所述目标图像中对应的包围框；
    基于所述包围框的顶点确定所述第一目标对象在所述目标图像中的第一二维几何特征。
  5. 根据权利要求3或4所述的定位方法，其特征在于，在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，所述基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征，包括：
    基于所述语义分割结果，从所述目标图像中确定属于所述第一目标对象的轮廓的像素点；
    基于属于所述第一目标对象的轮廓的像素点在所述目标图像中的位置，得到所述第一目标对象的轮廓线；
    基于所述第一目标对象的轮廓线，确定所述第一目标对象在所述目标图像中的第一二维几何特征。
  6. 根据权利要求3-5任一项所述的定位方法，其特征在于，在所述第一目标对象包括线型的第三道路标志对象的情况下，所述基于多个像素点分别对应的语义分割结果，以及所述多个像素点分别在所述目标图像中的位置，确定所述第一目标对象在所述目标图像中的第一二维几何特征，包括：
    基于所述语义分割结果，拟合得到第一目标对象的中心线；
    基于位于所述中心线上、且属于所述第一目标对象所在图像区域的像素点在所述目标图像中的二维坐标值，确定属于所述第一目标对象所在图像区域、且位于所述中心线上的目标线段；
    基于所述目标线段得到所述第一目标对象的第一二维几何特征。
  7. 根据权利要求1-6任一项所述的定位方法，其特征在于，还包括：基于所述目标场景中的第二目标对象在所述目标场景中的三维几何特征、所述第一目标对象的第一二维几何特征，生成所述第一目标对象和所述第二目标对象的所述对应关系。
  8. 根据权利要求7所述的定位方法，其特征在于，所述基于所述目标场景中的第二目标对象在所述目标场景中的三维几何特征、所述第一目标对象的第一二维几何特征，生成所述第一目标对象和所述第二目标对象的所述对应关系，包括：
    基于所述采集所述目标图像的采集设备的初始位姿信息、以及所述第二目标对象在所述目标场景中的三维几何特征，将所述第二目标对象投影至所述目标图像的图像坐标系中，得到所述第二目标对象在所述图像坐标系中的第一投影几何特征；
    基于所述第一目标对象在所述图像坐标系中的第一二维几何特征、以及所述第二目标对象在所述图像坐标系中的第一投影几何特征，对所述第一目标对象和所述第二目标对象进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系。
  9. 根据权利要求7或8所述的定位方法，其特征在于，所述基于所述目标场景中的第二目标对象在所述目标场景中的三维几何特征、所述第一目标对象的第一二维几何特征，生成所述第一目标对象和所述第二目标对象的所述对应关系，包括：
    基于所述目标图像与所述第二目标对象所在目标平面之间的单应矩阵，将所述目标图像中的第一目标对象投影至所述目标平面中，得到所述第一目标对象在所述目标平面中的第二投影几何特征；
    基于所述第一目标对象在所述目标平面中的第二投影几何特征、以及所述第二目标对象在所述目标平面中的几何特征，对所述第一目标对象和所述第二目标对象进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系；
    其中，所述第二目标对象在所述目标平面中的几何特征，是基于所述第二目标对象在所述场景坐标系中的三维几何特征确定的。
  10. 根据权利要求9所述的定位方法，其特征在于，所述基于所述第一目标对象在所述目标平面中的第二投影几何特征、以及所述第二目标对象在所述目标平面中的几何特征，对所述第一目标对象和所述第二目标对象进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系，包括：
    在所述第一目标对象包括具有矩形轮廓的第一道路标志对象的情况下，利用所述第二投影几何特征中表征所述第一目标对象的顶点的特征、以及所述第二目标对象在所述目标平面中的几何特征中表征所述第二目标对象确定的顶点的特征进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系；
    在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，利用所述第二投影几何特征中表征所述第一目标对象的轮廓线和/或角点的特征、以及所述第二目标对象在所述目标平面中的几何特征中表征所述第二目标对象确定的轮廓线和/或角点的特征进行匹配，得到所述第一目标对象和所述第二目标对象的对应关系；
    在第一目标对象包括线型的第三道路标志对象的情况下，对所述第二投影几何特征、以及所述第二目标对象在所述目标平面中的几何特征进行最大图匹配，得到所述第一目标对象和所述第二目标对象的对应关系。
  11. 根据权利要求1-10任一项所述的定位方法，其特征在于，所述基于位置匹配的结果，确定采集所述目标图像的采集设备的目标位姿信息，包括：基于位置匹配的结果，确定位置匹配误差；
    基于所述位置匹配误差以及采集所述目标图像的采集设备的初始位姿信息，确定采集所述目标图像的采集设备的目标位姿信息。
  12. 根据权利要求11所述的定位方法，其特征在于，所述基于所述位置匹配误差以及采集所述目标图像的采集设备的初始位姿信息，确定采集所述目标图像的采集设备的目标位姿信息，包括：
    检测是否满足预设的迭代停止条件；
    在满足所述迭代停止条件的情况下，将最后一次迭代得到的所述初始位姿信息，确定为所述目标位姿信息；
    在不满足所述迭代停止条件的情况下，基于所述位置匹配误差、以及最近一次迭代过程中的初始位姿信息，确定新的初始位姿信息，并返回至基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配的步骤。
  13. 根据权利要求12所述的定位方法，其特征在于，所述迭代停止条件包括下述任一项：
    迭代次数大于预设迭代次数阈值；
    所述第一目标对象和所述第二目标对象的位置匹配误差小于预设的损失阈值。
  14. 根据权利要求11-13任一项所述的定位方法，其特征在于，所述基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配，包括：
    在所述第一目标对象包括具有不规则轮廓的第二道路标志对象的情况下，对所述第一目标对象的第一二维几何特征进行插值处理，得到所述第一目标对象的第二二维几何特征；其中，所述第二二维几何特征包括：多个顶点、以及多个插值点；
    基于所述第二二维几何特征、所述三维几何特征以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行点对点的位置匹配。
  15. 根据权利要求1-14任一项所述的定位方法，其特征在于，所述基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配，包括：
    基于采集所述目标图像的所述采集设备的初始位姿信息、以及所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征，将所述第二目标对象投影至所述目标图像的图像坐标系中，得到所述第二目标对象在所述图像坐标系中的第三投影几何特征；
    基于所述第二目标对象在所述图像坐标系中的所述第三投影几何特征、所述第一目标对象的第一二维几何特征，对具有对应关系的第一目标对象和第二目标对象进行位置匹配。
  16. 一种智能行驶装置的行驶控制方法，其特征在于，包括：
    获取智能行驶装置在行驶过程中采集的视频帧数据；
    利用基于权利要求1-15任一项所述的定位方法处理所述视频帧数据，检测所述视频帧数据中的目标对象；
    基于检测的目标对象，控制所述智能行驶装置。
  17. 一种定位装置，其特征在于，包括：
    第一获取模块，用于获取对目标场景进行采集得到的目标图像；
    第一确定模块，用于基于所述目标图像，确定所述目标图像中包括的第一目标对象在所述目标图像中的第一二维几何特征；
    匹配模块，用于基于所述目标场景中的第二目标对象在所述目标场景对应的场景坐标系中的三维几何特征、所述第一目标对象的第一二维几何特征、以及所述第一目标对象和所述第二目标对象的对应关系，对所述第一目标对象和所述第二目标对象进行位置匹配；
    第二确定模块，用于基于位置匹配的结果，确定采集所述目标图像的采集设备的目标位姿信息。
  18. 一种智能行驶装置的行驶控制装置，其特征在于，包括：
    第二获取模块，用于获取智能行驶装置在行驶过程中采集的视频帧数据；
    检测模块，用于利用基于权利要求1-15任一项所述的定位方法处理所述视频帧数据，检测所述视频帧数据中的目标对象；
    控制模块，用于基于检测的目标对象，控制所述智能行驶装置。
  19. 一种计算机设备，其特征在于，包括：处理器、存储器，所述存储器存储有所述处理器可执行的机器可读指令，所述处理器用于执行所述存储器中存储的机器可读指令，所述机器可读指令被所述处理器执行时，所述处理器执行如权利要求1至15任一项所述的定位方法的步骤；或者执行如权利要求16所述的智能行驶装置的行驶控制方法的步骤。
  20. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质上存储有计算机程序，所述计算机程序被计算机设备运行时，所述计算机设备执行如权利要求1至15任一项所述的定位方法的步骤；或者执行如权利要求16所述的智能行驶装置的行驶控制方法的步骤。
  21. 一种计算机程序，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行用于实现权利要求1-15中的任一项权利要求所述的定位方法的步骤；或者执行如权利要求16所述的行驶控制方法的步骤。
PCT/CN2021/098964 2021-03-31 2021-06-08 定位方法、行驶控制方法、装置、计算机设备及存储介质 WO2022205618A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110349351.3 2021-03-31
CN202110349351.3A CN113095184B (zh) 2021-03-31 2021-03-31 定位方法、行驶控制方法、装置、计算机设备及存储介质

Publications (1)

Publication Number Publication Date
WO2022205618A1 true WO2022205618A1 (zh) 2022-10-06

Family

ID=76672135

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/098964 WO2022205618A1 (zh) 2021-03-31 2021-06-08 定位方法、行驶控制方法、装置、计算机设备及存储介质

Country Status (2)

Country Link
CN (1) CN113095184B (zh)
WO (1) WO2022205618A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115294204A (zh) * 2022-10-10 2022-11-04 浙江光珀智能科技有限公司 一种户外目标定位方法及系统
CN115775325A (zh) * 2023-01-29 2023-03-10 摩尔线程智能科技(北京)有限责任公司 一种位姿确定方法及装置、电子设备和存储介质

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116007637B (zh) * 2023-03-27 2023-05-30 北京集度科技有限公司 定位装置、方法、车载设备、车辆、及计算机程序产品

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106940186A (zh) * 2017-02-16 2017-07-11 华中科技大学 一种机器人自主定位与导航方法及系统
CN109284681A (zh) * 2018-08-20 2019-01-29 北京市商汤科技开发有限公司 位姿检测方法及装置、电子设备和存储介质
US20190129170A1 (en) * 2017-10-31 2019-05-02 Panasonic Intellectual Property Management Co., Lt d. Display system and movable object
CN110146096A (zh) * 2018-10-24 2019-08-20 北京初速度科技有限公司 一种基于图像感知的车辆定位方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109242913B (zh) * 2018-09-07 2020-11-10 百度在线网络技术（北京）有限公司 采集器相对参数的标定方法、装置、设备和介质
CN109887087B (zh) * 2019-02-22 2021-02-19 广州小鹏汽车科技有限公司 一种车辆的slam建图方法及系统
CN111862146B (zh) * 2019-04-30 2023-08-29 北京魔门塔科技有限公司 一种目标对象的定位方法及装置
CN111627001B (zh) * 2020-05-25 2024-05-17 深圳市商汤科技有限公司 图像检测方法及装置
CN111862199B (zh) * 2020-06-17 2024-01-09 北京百度网讯科技有限公司 定位方法、装置、电子设备和存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106940186A (zh) * 2017-02-16 2017-07-11 华中科技大学 一种机器人自主定位与导航方法及系统
US20190129170A1 (en) * 2017-10-31 2019-05-02 Panasonic Intellectual Property Management Co., Lt d. Display system and movable object
CN109284681A (zh) * 2018-08-20 2019-01-29 北京市商汤科技开发有限公司 位姿检测方法及装置、电子设备和存储介质
CN110146096A (zh) * 2018-10-24 2019-08-20 北京初速度科技有限公司 一种基于图像感知的车辆定位方法及装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115294204A (zh) * 2022-10-10 2022-11-04 浙江光珀智能科技有限公司 一种户外目标定位方法及系统
CN115775325A (zh) * 2023-01-29 2023-03-10 摩尔线程智能科技(北京)有限责任公司 一种位姿确定方法及装置、电子设备和存储介质

Also Published As

Publication number Publication date
CN113095184A (zh) 2021-07-09
CN113095184B (zh) 2023-01-31

Similar Documents

Publication Publication Date Title
WO2022205618A1 (zh) 定位方法、行驶控制方法、装置、计算机设备及存储介质
US11320833B2 (en) Data processing method, apparatus and terminal
EP3779358B1 (en) Map element extraction method and apparatus
US10824862B2 (en) Three-dimensional object detection for autonomous robotic systems using image proposals
KR102267562B1 (ko) 무인자동주차 기능 지원을 위한 장애물 및 주차구획 인식 장치 및 그 방법
US11670087B2 (en) Training data generating method for image processing, image processing method, and devices thereof
Pfeiffer et al. Towards a Global Optimal Multi-Layer Stixel Representation of Dense 3D Data.
CN113744560B (zh) 停车场自动泊车方法、装置、服务器和机器可读存储介质
WO2017041396A1 (zh) 一种车道线数据的处理方法、装置、存储介质及设备
US11120280B2 (en) Geometry-aware instance segmentation in stereo image capture processes
CA2950791A1 (en) Binocular visual navigation system and method based on power robot
CN111209770A (zh) 一种车道线识别方法及装置
Held et al. A probabilistic framework for car detection in images using context and scale
CN116993817B (zh) 目标车辆的位姿确定方法、装置、计算机设备及存储介质
JP2022542082A (ja) ポーズ特定方法、ポーズ特定装置、コンピュータ可読記憶媒体、コンピュータ機器及びコンピュータプログラム
CN113420648B (zh) 一种具有旋转适应性的目标检测方法及系统
Börcs et al. A model-based approach for fast vehicle detection in continuously streamed urban LIDAR point clouds
Chun-Zhao et al. Drivable road boundary detection for intelligent vehicles based on stereovision with plane-induced homography
Huang et al. A coarse-to-fine LiDar-based SLAM with dynamic object removal in dense urban areas
JP2023146041A (ja) 機械学習システム
He et al. AutoMatch: Leveraging Traffic Camera to Improve Perception and Localization of Autonomous Vehicles
Zhang et al. Dijkstra model for stereo-vision based road detection: a non-parametric method
CN113189610A (zh) 地图增强的自动驾驶多目标追踪方法和相关设备
Salah et al. Summarizing large scale 3D mesh for urban navigation
Wei et al. Plane-based scan registration with moving vehicles exclusion

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21934267

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE