WO2022213867A1 - 目标物识别 - Google Patents

目标物识别 Download PDF

Info

Publication number
WO2022213867A1
WO2022213867A1 PCT/CN2022/084308 CN2022084308W WO2022213867A1 WO 2022213867 A1 WO2022213867 A1 WO 2022213867A1 CN 2022084308 W CN2022084308 W CN 2022084308W WO 2022213867 A1 WO2022213867 A1 WO 2022213867A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
image
recognition result
target object
recognized
Prior art date
Application number
PCT/CN2022/084308
Other languages
English (en)
French (fr)
Inventor
夏华夏
乔健
Original Assignee
北京三快在线科技有限公司
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 北京三快在线科技有限公司 filed Critical 北京三快在线科技有限公司
Publication of WO2022213867A1 publication Critical patent/WO2022213867A1/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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
    • 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
    • G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
    • G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Definitions

  • the present disclosure relates to the field of unmanned driving technology, and in particular, to target object recognition.
  • the control of unmanned equipment relies on the recognition of objects in the surrounding environment, so as to control the driving of unmanned equipment based on the recognition results.
  • the target usually refers to: obstacles, signs, signal lights and other objects that will affect the driving of the unmanned vehicle.
  • Embodiments of the present disclosure provide a target object identification method and apparatus to partially solve the above-mentioned problems in the prior art.
  • a target object recognition method includes: from a collected image, determining a first area corresponding to each target object; The images of the first regions corresponding to the objects respectively, determine the images to be recognized; input the images to be recognized into the trained target detection model, and determine the second corresponding images of the objects in the images to be recognized.
  • the images of the second regions corresponding to each of the targets in the image are input into the classification model after training, and the second recognition result of each of the targets is determined; according to the first recognition result and the second recognition result of each of the targets , to determine the final recognition result of each of the targets.
  • the present disclosure provides a target object identification device, comprising: a determination module that determines the corresponding corresponding objects in the collected images according to the pre-stored positions of the target objects and the pose of the collection device when the images are collected.
  • the extraction module extracts the image of the first region corresponding to each target in the image, and determines the image to be recognized;
  • the first recognition module inputs the to-be-recognized image into the trained target detection model, Determine the corresponding second area of each of the targets in the to-be-recognized image, and the first recognition result of each of the targets, wherein, for each of the targets, the range of the second area corresponding to the target is smaller than the range of the first area corresponding to the target;
  • the second recognition module inputs the images of the second area corresponding to each target in the to-be-recognized image into the trained classification model, and determines the The second recognition result;
  • the recognition module determines the final recognition result of each target object according to the first recognition result and the second recognition
  • the present disclosure provides a computer-readable storage medium, where a computer program is stored in the storage medium, and when the computer program is executed by a processor, the above-mentioned target object identification method is implemented.
  • the present disclosure provides an unmanned vehicle, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, the processor implements the above-mentioned target when the processor executes the program recognition methods.
  • the first area of each target object is determined from the collected images, and the to-be-recognized image is determined according to the image of the first area of each target object Image, through the target detection model, determine the second area of each target in the to-be-recognized image and the first recognition result, through the classification model, from the image of the second area of each target, determine the second recognition of each target As a result, the final recognition result of each target is determined according to the first recognition result and the second recognition result of each target.
  • this method can reduce the problem of the reduction of recognition accuracy caused by noise by extracting the corresponding area of the target object in the collected image.
  • the corresponding area, and the recognition result of each target, and finally through the two-stage recognition result the final recognition result is determined, and the accuracy of target recognition is improved.
  • FIG. 1 is a schematic flowchart of a target object identification method provided by the present disclosure
  • FIG. 2 is a schematic diagram of determining an image to be recognized provided by the present disclosure
  • FIG. 3 is a schematic diagram of determining a second region provided by the present disclosure.
  • FIG. 4 is a schematic diagram of determining a first identification result provided by the present disclosure
  • FIG. 5 is a schematic diagram of determining a second identification result provided by the present disclosure.
  • FIG. 6 is a schematic diagram of determining a first region corresponding to a target provided by the present disclosure
  • FIG. 7 is a schematic structural diagram of a target object identification device provided by the present disclosure.
  • FIG. 8 is a schematic structural diagram of the unmanned device corresponding to FIG. 1 provided in the present disclosure.
  • the control of unmanned equipment relies on the recognition of objects in the surrounding environment, so as to control the driving of unmanned equipment based on the recognition results.
  • the target usually refers to: obstacles, signs, signal lights and other objects that will affect the driving of the unmanned vehicle.
  • the identification of the signal light is mainly based on the high-precision map.
  • the area where each signal light is located is roughly determined in the collected image, and then determined.
  • the image of the area where each signal light is located (for example, the minimum circumscribed rectangle of each signal light is determined in the collected image), the determined image is input into the pre-trained signal light recognition model, and the status of each signal light output by the signal light recognition model is obtained as Identify the results.
  • the state of the signal light includes, for example, a red light, a yellow light, a green light, or an off state.
  • the image of the area where each signal light is located as the input can reduce the amount of calculation, so that the identification result of each signal light can be determined by only one calculation.
  • the image of the area where each signal light is located may still contain a lot of noise, for example, the background image between the signal lights, other obstacles on the road, or the lights of the shops on the street, etc., which may easily interfere with the identification of the signal lights.
  • the signal light is far away, the size of the signal light in the image is small, and it is generally difficult to detect small targets, so the phenomenon of missed detection is easy to occur.
  • FIG. 1 is a schematic flowchart of a target object identification method provided by the present disclosure, including:
  • S100 From the collected images, determine the first regions corresponding to the respective objects.
  • the unmanned device can continuously collect images around itself or in the forward direction through the acquisition device set on the unmanned device. And when the target needs to be identified, the image collected at the current moment is determined. Therefore, the target objects around the unmanned device can be recognized based on the image collected at the current moment, and then the driving of the unmanned device can be controlled based on the recognition result, so as to ensure the safe driving of the unmanned device.
  • the unmanned device Under normal circumstances, after the current image is collected, the unmanned device itself can recognize the image, determine the recognition results of each target, and then determine its own motion strategy at the next moment. For the convenience of description, the following description will be given by taking the method for recognizing the target object performed by an unmanned device as an example.
  • the unmanned device may determine the first area corresponding to each target from the collected image.
  • each target can be an object such as an obstacle, a sign, a signal light, etc., which will affect the driving of the unmanned device.
  • a signal light is used as an example for description in the following.
  • a high-precision map is usually pre-stored, so that the unmanned device can determine its own position and other information according to the high-precision map and the image collected by the acquisition device, and based on the determined information such as its own position Determine exercise strategy. Therefore, after determining the collected image, the unmanned vehicle can first determine, from the pre-stored high-precision map, according to the pose of the collection device when the image was collected, the targets within the collection range when the collection device collects the image. thing.
  • the pose is used to represent information such as orientation, acceleration, and steering when the unmanned device collects the image.
  • the acquisition range that is, the image acquisition range corresponding to the unmanned device in the acquisition pose.
  • a high-precision map is a map in which the position information of each target is accurately marked for the convenience of driving of unmanned vehicles. That is to say, as long as the map with the position information of each target is marked, it can be used as the high-precision map in the present disclosure.
  • the unmanned device can determine the corresponding area of each target from the image according to the determined position of each target and the attributes of each target, as The first area corresponding to each target.
  • the attributes of each target may include the type, size, and shape of the target.
  • the shape of the signal light is a rectangle
  • the shape of the traffic sign can be a circle, a triangle, a rectangle, and the like.
  • an unmanned device may refer to a device capable of realizing automatic driving, such as an unmanned vehicle, a robot, and an automatic distribution device. Based on this, the unmanned device applying the target recognition method provided by the present disclosure can be used to perform distribution tasks in the field of distribution, such as the use of the unmanned device for express delivery, logistics, takeaway and other business scenarios.
  • the safe driving of human-driven equipment in the delivery business scenario requires the identification of static objects with specific traffic indication functions such as traffic lights and traffic signs set on the road through the collected images.
  • S102 Determine an image to be recognized according to the image of the first region corresponding to each target in the image.
  • the unmanned vehicle may determine the image according to the image of the first area corresponding to each target in the image. Image to be recognized.
  • the unmanned vehicle may extract sub-images of each target from the image according to the first regions corresponding to each target determined in step S100, splicing the sub-images of each target, and The acquired image after splicing is used as the image to be recognized, as shown in FIG. 2 .
  • FIG. 2 is a schematic diagram of determining an image to be recognized provided by the present disclosure.
  • the left part of FIG. 2 is an image collected when an unmanned device is located at an intersection. It can be seen that there are multiple signal lights in the image, so the unmanned device Based on the high-precision map, vehicle pose, image sensor pose, etc., the approximate position of each signal light can be determined, that is, the areas A, B, C, D, E in the image, and the areas A, B, C, D, E, as the first area corresponding to each signal lamp. Then the unmanned device can determine the image to be recognized in the right part of Fig. 2 according to the images corresponding to the areas A, B, C, D, and E in the left part.
  • the image to be recognized is determined by the No.
  • the image composition of an area area A corresponds to 1 part of the image to be recognized, area B corresponds to 2 parts of the image to be recognized, area C corresponds to 3 parts of the image to be recognized, area D corresponds to 4 parts of the image to be recognized, and area E is to be recognized Figure corresponds to the 5 parts.
  • the image to be recognized can usually be set as a rectangle, and different splicing rules may cause the result of splicing the sub-images of each target to be non-rectangular, as shown in Figure 2 by 1, 2, 3, 4, and 5-part polygons. Therefore, in the present disclosure, the unmanned device can also complete the image for each non-rectangular mosaic result.
  • the specific method for determining the image to be recognized can be set as required, which is not limited in the present disclosure.
  • S104 Input the to-be-recognized image into the trained target detection model, and determine the second area corresponding to each target in the to-be-recognized image, and the first recognition result of each target, wherein for each of the A target, the range of the second area corresponding to the target is smaller than the range of the first area corresponding to the target.
  • the unmanned device may input the to-be-recognized image into a trained target detection model, and determine that each target is in the to-be-recognized image. and the first recognition result of each target.
  • the first area determined in step S100 is the approximate range of the target. Therefore, the determined image to be recognized based on the images of the first regions corresponding to each target object also defines the approximate range of each target object. Therefore, after determining the to-be-recognized image, the unmanned vehicle can determine the specific position of each target in the to-be-recognized image according to the to-be-recognized image and the pre-trained target detection model, as each target in the to-be-recognized image.
  • the corresponding second area in the image to be recognized is shown in FIG. 3 .
  • FIG. 3 is a schematic diagram of determining the second area provided by the present disclosure
  • the left part in FIG. 3 is the image to be recognized determined in step S102
  • the middle part is the target detection model
  • the right part is the target object in the to-be-recognized image respectively.
  • the corresponding second area it can be seen that the second area corresponding to each target is the specific range of each target determined based on the image to be recognized. Therefore, for each target, the second area corresponding to the target is The range is smaller than the range of the first region.
  • the unmanned device while determining the second area corresponding to each target, can also determine the type of the target in the to-be-recognized image through a target detection model , as the first recognition result of each target, as shown in FIG. 4 .
  • FIG. 4 is a schematic diagram of determining the first recognition result provided by the present disclosure.
  • the left part in FIG. 4 is the image to be recognized determined in step S102 , assuming that the signal lights are yellow, red, and green respectively from top to bottom, and white means off. , black represents the light, then the signal light state corresponding to the sub-image of each target is green, the middle part in Figure 4 is the target detection model, and the right part is the first recognition result corresponding to each target in the image to be recognized.
  • the first identification result is the probability that the signal light state corresponding to each target is off, the signal light state is red, the signal light state is yellow, and the signal light state is green, respectively. It can be seen that in the first identification result corresponding to each target, the probability of the signal light state being green is relatively high.
  • S106 Input the images of the second regions corresponding to the respective targets into the trained classification model, and determine the second recognition result of each target.
  • the unmanned vehicle can also be determined according to step S104.
  • the second area corresponding to each target object that is, the accurate range of each target object, determine the image of the second range corresponding to each target object, and input the image of the second area corresponding to each target object to the classification after training.
  • the model determines the second recognition result of each target.
  • the unmanned device can extract the image of the second area corresponding to the target from the image to be recognized according to the second area corresponding to the target determined in step S104, and Input the extracted image of the second region corresponding to the target into the trained classification model to determine the second recognition result of the target, as shown in FIG. 5 .
  • FIG. 5 is a schematic diagram of determining the second recognition result provided by the present disclosure.
  • the left part of FIG. 5 is an image of the second area corresponding to the target. Similar to FIG. 4 , it is assumed that the signal lights are yellow, red, Green light, white means off, black means on, then the signal light state corresponding to the image of the second area corresponding to the target is green, the middle part is the classification model, and the right part is the second recognition result, similar to the first recognition result, the first
  • the second recognition result is the probability that the signal light state corresponding to each target is off, the signal light state is red, the signal light state is yellow, and the signal light state is green, respectively. It can be seen that in the second recognition result corresponding to the image of the second area corresponding to the target object, the probability that the status of the signal light is green is relatively high.
  • the first identification result and the second identification result in the present disclosure can indicate whether each target is a preset classification, such as whether the status of the signal light is red, whether the status of the signal light is off, etc., and can also indicate whether each target is in a preset classification.
  • the probability that the object belongs to each preset classification such as the probability that the signal light state is red, the probability that the signal light state is green, etc., the specific content of the first recognition result and the second recognition result can be set as required, which is not covered in this disclosure. make restrictions.
  • S108 Determine the final recognition result of each target according to the first recognition result and the second recognition result of each target.
  • the unmanned device may also use the first recognition result and the second recognition result of each target. Identify the results to determine the final identification results of each target.
  • the first recognition result of each target determined in step S104 and the second recognition result of each target determined in step S106 may be inaccurate in recognition.
  • the final recognition result of each target is determined by the first recognition result and the second recognition result of each target, and the weight of each recognition result, a more accurate recognition result can be obtained.
  • the weight of the first recognition result and the weight of the second recognition result of each target can be preset, and when the target needs to be recognized, for each target, according to the first recognition result of the target and its weight
  • the first weight, the second recognition result and its second weight, the first recognition result and its first weight of the target object, and the second recognition result and its second weight are input into a predetermined weighting determination function
  • X 1 is the first recognition result corresponding to the target
  • a 1 is the first weight corresponding to the first recognition result
  • X 2 is the second recognition result corresponding to the target
  • a 2 is the second recognition result corresponding second weight
  • the weighted result of the target object can be obtained
  • the weighted result is the weighted average probability of the target object belonging to each preset classification respectively.
  • the unmanned vehicle may determine the state with the highest probability from the determined weighted average probability that the target object belongs to each preset classification, as The final recognition result of the target. For example, if the weighted average probabilities of the corresponding states of the signal lights being off, red, yellow, and green are 10%, 80%, 2%, and 8%, respectively, it can be determined that the signal light state corresponding to the final recognition result of the target object is red.
  • the first area of each target is determined from the collected images, the image to be recognized is determined according to the image of the first area of each target, and the target detection model is used to determine the image to be recognized.
  • the second area and the first recognition result of each target are determined from the images of the second area of each target through the classification model, and the second recognition result of each target is determined according to the first recognition result and the first recognition result of each target 2.
  • Identification result determine the final identification result of each target.
  • step S100 the position of the target determined according to the high-precision map is only the approximate position of the target in the image. Therefore, in order to ensure the accuracy of the determined first area, the unmanned vehicle
  • the area corresponding to each target object may also be enlarged according to the value of the amplification parameter, and the enlarged area may be used as the first area corresponding to each target object. As shown in Figure 6.
  • the location of the target can be roughly determined according to the high-precision map
  • the solid line frame is the first area of the target determined by the original size of the target. It can be seen that, according to the first area of the target object determined by the original size of the target object, the recognition result of the target object cannot be accurately determined. Therefore, the unmanned device can enlarge the area corresponding to the target object according to the amplification parameter value, and the dotted frame is the first area of the target object determined according to the amplification parameter value. Obviously, according to the enlarged image of the first region, the recognition result of the target can be accurately determined in the subsequent steps. Wherein, the original size of the target object is determined according to the current pose of the unmanned device, a high-precision map, etc., and the value of the amplification parameter may be preset.
  • step S100 when the first area is determined according to the amplification parameter value, for different objects, the amplification parameter value is not exactly the same. If it is small, it may cause that the determined first area does not completely contain the target object, resulting in an error in the recognition result. For a target that is farther away from the unmanned device, if the value of the amplification parameter is too large, it may cause too much content of non-target objects in the determined first area, resulting in an error in the recognition result.
  • the unmanned device when the unmanned device amplifies the area corresponding to each target according to the amplification parameter value, it can also for each target, according to the position of the target, the properties of the target, and the The pose of the device is collected, and the magnification parameter value corresponding to the target is determined.
  • the distance between the target and the unmanned vehicle is determined according to the position of the target and the pose of the acquisition device when the image was collected, and then the corresponding amplification parameter value of the target is determined according to the distance, where the distance negatively correlated with the value of this amplification parameter. That is to say, the closer the target is to the unmanned equipment, the larger the value of the amplification parameter, and the farther the target is from the unmanned equipment, the smaller the value of the amplification parameter, so that the target object with a shorter distance corresponds to the larger value of the amplification parameter.
  • the target object is completely contained in the first area, and the first area of the distant target object contains less content of non-target objects, thereby avoiding the situation that the recognition result is erroneous due to inaccurate images to be recognized.
  • the corresponding magnification parameter value of the target is determined, wherein the size of the target is positively correlated with the magnification parameter value. That is to say, in the same image, when the first area is determined according to the same magnification parameter value, it may appear that the first area corresponding to the smaller-sized target completely contains the target, but the first area corresponding to the larger-sized target may appear. In the case where an area has not completely contained the target, therefore, the larger the target, the larger the magnification parameter value of the target when the first area is determined.
  • magnification parameter value can also be determined according to other information in the attributes of the target object, such as the shape of the target object.
  • the specific way of determining the amplification parameter value can be set as required, which is not limited in the present disclosure.
  • the attributes of each target can also include traffic information, and the specific attributes of each target can also be set as required, which is not limited in the present disclosure, wherein the traffic information can be obtained through server delivery or active query. .
  • the unmanned device can also determine, according to the first area and the second area of each target, in the images collected by the unmanned device, each The location of the target is marked and marked. Therefore, for each target, the unmanned device may determine the position of the target in the image of the corresponding first area according to the position of the target in the corresponding second area in the image to be recognized.
  • the unmanned device can also determine the position of the target object in the corresponding first area in the collected image and the position of the target object in the corresponding first area.
  • the position in the image of the first region determines the position of the target in the acquired image.
  • the unmanned device may also regard the sub-image of each target object as a separate image to be recognized, and continue to perform subsequent steps, so that the sub-image of each target object can be separately
  • the recognition image is processed to determine the final recognition result of each target separately.
  • the target object recognition method can also be executed by a server.
  • the unmanned device can also send the image to the server that detects the target object.
  • the server recognizes the target in the image according to the received image, and sends the recognition result to the unmanned device or the server that determines the control strategy of the unmanned device, and determines the movement of the unmanned device at the next moment. Strategy.
  • the target object recognition method provided by the present disclosure can be applied to determine the position and type of each target in the environmental image of the unmanned equipment when determining the motion strategy of the unmanned equipment.
  • the unmanned device can determine the motion strategy of the unmanned device at the next moment based on the determined location and type of the target, so that the unmanned device can drive normally without traffic at the next moment.
  • ACCIDENT The specific method of determining the motion strategy of the unmanned device according to the type and position of the target is already a relatively mature technology, and this disclosure will not describe it again.
  • the target detection model in step S104 can be obtained by pre-training by the server for training the model.
  • the to-be-recognized image determined by the first region corresponding to each target object determined by using several images historically collected by the acquisition device can be obtained as each training sample, and each target in each training sample can be obtained.
  • the object is labeled, and the location and type of the target in each training sample are used as training labels.
  • the server can input each training sample into the target detection model to be trained, determine the detection result of each target, and take minimizing the difference between the detection result of each training sample and the sample label as the optimization goal, and the target The detection model is trained.
  • the classification model in step S106 can be obtained by pre-training by the server for training the model.
  • training the classification model several labeled target images can be obtained as training samples.
  • Input into the classification model to be trained determine the classification result of each training sample, and train the classification model with the optimization goal of minimizing the difference between the classification result of each training sample and the annotation.
  • an embodiment of the present disclosure also provides a schematic structural diagram of a target object recognition device, as shown in FIG. 7 .
  • FIG. 7 is a schematic structural diagram of a target object identification device provided by an embodiment of the present disclosure, and the device includes:
  • a determination module 200 configured to respectively determine the first area corresponding to each target in the captured image according to the pre-stored position of the target and the pose of the capture device when capturing the image;
  • an extraction module 202 configured to extract an image of the first region corresponding to each of the objects in the image, and determine an image to be recognized
  • the first recognition module 204 is used to input the image to be recognized into the trained target detection model, to determine the corresponding second area of each target in the to-be-recognized image, and the first area of each target. an identification result, wherein, for each of the targets, the range of the second area corresponding to the target is smaller than the range of the first area corresponding to the target;
  • the second recognition module 206 is configured to input the image of the second region corresponding to each target object in the to-be-recognized image into the trained classification model, and determine the second recognition result of each of the target objects;
  • the identification module 208 is configured to determine the final identification result of each target object according to the first identification result and the second identification result of each target object.
  • the determining module 200 is specifically configured to determine, in the pre-stored high-precision map, according to the pose of the acquisition device when the image is acquired, the location of each location within the acquisition range when the acquisition device acquires the image. According to the position of each target object and the attribute of each target object, the area corresponding to each target object is determined in the collected image respectively, and according to the value of the magnification parameter, each target object is respectively The region corresponding to the object is enlarged, and the enlarged region is used as the first region corresponding to each target object.
  • the extraction module 202 is specifically configured to extract sub-images of each of the targets from the image according to the first regions corresponding to each of the targets, and to extract the sub-images of each of the targets from the image.
  • the image obtained after splicing the sub-images is used as the to-be-identified image.
  • the second recognition module 206 is specifically configured to, for each target, extract the target from the to-be-recognized image according to the second region corresponding to the target output by the target detection model.
  • the image of the second area corresponding to the target is input into the trained classification model to determine the second recognition result of the target.
  • the first recognition result and the second recognition result of each of the targets are the probability that each of the targets belongs to each preset classification.
  • the recognition module 208 is specifically configured to, for each of the targets, According to the first recognition result, the second recognition result, the first weight and the second weight of the target object, determine the weighted average probability that the target object belongs to each of the preset classifications. Set the weighted average probability of classification to determine the final recognition result of the target.
  • the determining module 200 determines the zoom parameter corresponding to the target object according to the position of the target object, the attribute of the target object, and the pose of the acquisition device when the image is captured value, according to the corresponding zoom parameter value of the target, the area corresponding to the target is enlarged.
  • the identification module 208 is specifically configured to, for each of the target objects, determine that the target object is in the corresponding first region according to the position of the target object in the corresponding second region in the to-be-recognized image. According to the position of the target in the corresponding first area in the collected image, and the position of the target in the corresponding image of the first area, determine the target in the collected image The position in the image, according to the position of the target in the captured image, the target is marked in the captured image.
  • the present disclosure also provides a computer-readable storage medium, where the storage medium stores a computer program, and when the computer program is executed by the processor, can be used to execute the target object identification method provided above.
  • an embodiment of the present disclosure also provides a schematic structural diagram of the unmanned device shown in FIG. 8 .
  • the driverless device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and of course, it may also include hardware required by other services.
  • the processor reads the corresponding computer program from the non-volatile memory into the memory and executes it, so as to realize the above-mentioned target object identification method.
  • a Programmable Logic Device (such as a Field Programmable Gate Array (FPGA)) is an integrated circuit whose logic function is determined by user programming of the device.
  • HDL Hardware Description Language
  • ABEL Advanced Boolean Expression Language
  • AHDL Altera Hardware Description Language
  • HDCal Joint Component Interconnect
  • JHDL Java Hardware Description Language
  • Lava Lava
  • Lola MyHDL
  • PALASM RHDL
  • VHDL Very-High-Speed Integrated Circuit Hardware Description Language
  • Verilog Verilog
  • the controller may be implemented in any suitable manner, for example, the controller may take the form of eg a microprocessor or processor and a computer readable medium storing computer readable program code (eg software or firmware) executable by the (micro)processor , logic gates, switches, application specific integrated circuits (ASICs), programmable logic controllers and embedded microcontrollers, examples of controllers include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicon Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory.
  • the controller may take the form of eg a microprocessor or processor and a computer readable medium storing computer readable program code (eg software or firmware) executable by the (micro)processor , logic gates, switches, application specific integrated circuits (ASICs), programmable logic controllers and embedded microcontrollers
  • ASICs application specific integrated circuits
  • controllers include but are not limited to
  • the controller in addition to implementing the controller in the form of pure computer-readable program code, the controller can be implemented as logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded devices by logically programming the method steps.
  • the same function can be realized in the form of a microcontroller, etc. Therefore, this kind of controller can be regarded as a hardware component, and the devices included therein for realizing various functions can also be regarded as a structure in the hardware component. Or even, the means for implementing various functions can be regarded as both a software module implementing a method and a structure within a hardware component.
  • a typical implementation device is a computer.
  • the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or A combination of any of these devices.
  • embodiments of the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • computer-usable storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions
  • the apparatus implements the functions specified in the flow or flows of the flowcharts and/or the block or blocks of the block diagrams.
  • a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • processors CPUs
  • input/output interfaces network interfaces
  • memory volatile and non-volatile memory
  • Memory may include forms of non-persistent memory, random access memory (RAM) and/or non-volatile memory in computer readable media, such as read only memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
  • RAM random access memory
  • ROM read only memory
  • flash RAM flash memory
  • Computer-readable media includes both persistent and non-permanent, removable and non-removable media, and storage of information may be implemented by any method or technology.
  • Information may be computer readable instructions, data structures, modules of programs, or other data.
  • Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Flash Memory or other memory technology, Compact Disc Read Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cartridges, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.
  • computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.
  • embodiments of the present disclosure may be provided as a method, system or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • computer-usable storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.
  • program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types.
  • the disclosure may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network.
  • program modules may be located in both local and remote computer storage media including storage devices.

Abstract

本公开公开了一种目标物识别方法以及装置，通过对采集图像中目标物对应区域的提取，减少噪声引起的识别准确率下降的问题，同时通过两阶段的识别，可以进一步精准地确定每个目标物对应的区域，以及每个目标物的识别结果，最后通过两阶段识别结果，确定最终识别结果，提高对目标物识别准确率。

Description

目标物识别 技术领域
本公开涉及无人驾驶技术领域，尤其涉及目标物识别。
背景技术
无人驾驶设备的控制，依赖于对周围环境中目标物的识别，以基于识别结果控制无人驾驶设备行驶。其中，目标物通常是指：障碍物、指示牌、信号灯等会影响无人驾驶设备行驶的对象。
发明内容
本公开实施例提供一种目标物识别方法以及装置，以部分解决上述现有技术存在的问题。
本公开实施例采用下述技术方案：第一方面，本公开提供的一种目标物识别方法，包括：从采集的图像中，确定各目标物分别对应的第一区域；根据所述图像中各所述目标物分别对应的第一区域的图像，确定待识别图像；将所述待识别图像输入训练完成的目标检测模型，确定各所述目标物在所述待识别图像中分别对应的第二区域，以及各所述目标物的第一识别结果，其中，针对各所述目标物，该目标物对应的第二区域的范围小于该目标物对应的第一区域的范围；将所述待识别图像中各所述目标物分别对应的第二区域的图像输入训练完成的分类模型，确定各所述目标物的第二识别结果；根据各所述目标物的第一识别结果以及第二识别结果，确定各所述目标物的最终识别结果。
第二方面，本公开提供的一种目标物识别装置，包括：确定模块，根据预存的目标物的位置以及采集图像时的采集设备位姿，在采集的所述图像中分别确定各目标物对应的第一区域；提取模块，提取所述图像中各所述目标物对应的第一区域的图像，确定待识别图像；第一识别模块，将所述待识别图像输入训练完成的目标检测模型，确定各所述目标物在所述待识别图像中的对应的第二区域，以及各所述目标物的第一识别结果，其中，针对各所述目标物，该目标物对应的第二区域范围小于该目标物对应的第一区域的范围；第二识别模块，将所述待识别图像中各所述目标物对应的第二区域的图像输入训练完成的分类模型，确定各所述目标物的第二识别结果；识别模块，根据各所述目标物的第一识别结果以及第二识别结果，确定各所述目标物的最终识别结果。
第三方面，本公开提供的一种计算机可读存储介质，所述存储介质存储有计算机程序，所述计算机程序被处理器执行时实现上述的目标物识别方法。
第四方面，本公开提供的一种无人驾驶设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现上述的目标物识别方法。
本公开实施例采用的上述至少一个技术方案能够达到以下有益效果：在本公开中，从采集的图像中确定各目标物的第一区域，根据各目标物的第一区域的图像，确定待识别图像，通过目标检测模型，确定该待识别图像中各目标物的第二区域和第一识别结果，通过分类模型，从各目标物的第二区域的图像中，确定各目标物的第二识别结果，根据各目标物的第一识别结果以及第二识别结果，确定各目标物的最终识别结果。
从上述方法可以看出，本方法通过对采集图像中目标物对应区域的提取，减少噪声引起的识别准确率下降的问题，同时通过两阶段的识别，可以进一步精细化确定每个目标物对应的区域，以及每个目标物的识别结果，最后通过两阶段识别结果，确定最终识别结果，提高对目标物识别准确率。
附图说明
此处所说明的附图用来提供对本公开的进一步理解，构成本公开的一部分，本公开的示意性实施例及其说明用于解释本公开，并不构成对本公开的不当限定。在附图中：
图1为本公开提供的目标物识别方法的流程示意图；
图2为本公开提供的确定待识别图像的示意图；
图3为本公开提供的确定第二区域的示意图；
图4为本公开提供的确定第一识别结果的示意图；
图5为本公开提供的确定第二识别结果的示意图；
图6为本公开提供的确定目标物对应的第一区域的示意图；
图7为本公开提供的目标物识别装置的结构示意图；
图8为本公开提供的对应于图1的无人驾驶设备的结构示意图。
具体实施方式
为使本公开的目的、技术方案和优点更加清楚，下面将结合本公开具体实施例及相应的附图对本公开技术方案进行清楚、完整地描述。所描述的实施例仅是本公开一部分实施例，而不是全部的实施例。基于本公开中的实施例，本领域普通技术人员在没有做出创造性劳动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。
无人驾驶设备的控制，依赖于对周围环境中目标物的识别，以基于识别结果控制无人驾驶设备行驶。其中，目标物通常是指：障碍物、指示牌、信号灯等会影响无人驾驶设备行驶的对象。
以信号灯为例进行说明，对信号灯的识别主要是基于高精地图，通过实时确定出的车辆位姿，以及图像传感器位姿，在采集的图像中大致确定各信号灯所在的区域，之后，再确定各信号灯所在的区域的图像(如，在采集的图像中确定各信号灯的最小外接矩形)，将确定出的图像输入预先训练的信号灯识别模型，得到该信号灯识别模型输出的各信号灯的状态，作为识别结果。其中，信号灯的状态包括如：红灯、黄灯、绿灯或者是熄灭等状态。
将各信号灯所在的区域的图像作为输入，可以减少计算量，使得只需计算一次便可确定各信号灯的识别结果。但是，一般来说各信号灯所在的区域的图像中仍然可能包含大量噪声，例如，信号灯之间背景图像，路上的其他障碍物，或者临街店铺的灯光等等，容易对信号灯的识别产生干扰。并且，当信号灯距离较远时，信号灯在图像中的尺寸较小，而对小目标的检测一般较为困难，因此还容易产生漏检现象。
有鉴于此，以下结合附图，详细说明本公开各实施例提供的技术方案。
图1为本公开提供的目标物识别方法的流程示意图，包括：
S100：从采集的图像中，确定各目标物分别对应的第一区域。
一般的，在无人驾驶设备行驶过程中，周围环境会随着时间改变，因此，该无人驾驶设备可通过无人驾驶设备上设置的采集设备，连续采集自身四周或前进方向上的图像，并在需要对目标物进行识别时，确定当前时刻采集的图像。于是，可基于当前时刻采集到的图像，对无人驾驶设备周围的目标物进行识别，进而基于识别结果控制无人驾驶设备行驶，以保证无人驾驶设备的安全行驶。
通常情况下，在采集到当前图像后，可由无人驾驶设备本身对该图像进行识别，确 定各目标物的识别结果，进而确定下一时刻自身的运动策略。为了方便描述，后续以无人驾驶设备执行该目标物识别方法为例进行说明。
在本公开提供的一个或多个实施例中，在采集到当前图像后，该无人驾驶设备可从该采集到的图像中，确定各目标物分别对应的第一区域。其中，各目标物可为障碍物、指示牌、信号灯等会影响无人驾驶设备行驶的对象。为了方便描述，后续以信号灯为例进行说明。
具体的，在无人驾驶设备中，通常预存有高精地图，以使无人驾驶设备可根据高精地图以及采集设备采集到的图像确定自身位置等信息，并基于确定出的自身位置等信息确定运动策略。因此，在确定出采集的图像后，该无人驾驶设备可首先根据采集图像时的采集设备位姿，从预存的高精地图中，确定该采集设备采集该图像时，采集范围内的各目标物。其中，所述位姿用于表示无人驾驶设备采集该图像时的朝向、加速度、转向等信息。而由于图像采集器在无人驾驶设备上的位置是固定的，所以，一旦确定出无人驾驶设备采集该图像时的位姿，即可确定出该采集设备在采集该图像时所对应的图像采集范围，即无人驾驶设备处于采集位姿时所对应的图像采集范围。高精地图是为了便于无人驾驶设备行驶而准确地标注了各目标物的位置信息的地图。也就是说，只要标注出各目标物的位置信息的地图，都可作为本公开中的高精地图。
然后，在确定出该图像中的各目标物后，该无人驾驶设备可根据确定出的各目标物的位置以及各目标物的属性，从该图像中分别确定各目标物对应的区域，作为各目标物对应的第一区域。其中，各目标物的属性可包括目标物的种类、尺寸、以及形状等。如，信号灯的形状为矩形，交通指示牌的形状可为圆形、三角形、矩形等。
在本公开中，无人驾驶设备可以是指无人车、机器人、自动配送设备等能够实现自动驾驶的设备。基于此，应用本公开提供的目标识别的方法的无人驾驶设备可以用于执行配送领域的配送任务，如，使用无人驾驶设备进行快递、物流、外卖等配送的业务场景，而为了保证无人驾驶设备在配送业务场景中的安全行驶，需要通过采集到的图像，对道路中设置的交通灯、交通指示牌等具体交通指示功能的静态目标物进行识别。
S102：根据所述图像中各目标物分别对应的第一区域的图像，确定待识别图像。
在本公开提供的一个或多个实施例中，在确定出各目标物分别对应的第一区域后，该无人驾驶设备可根据该图像中各目标物分别对应的第一区域的图像，确定待识别图像。
具体的，该无人驾驶设备可根据步骤S100中确定出的各目标物分别对应的第一区 域，从该图像中提取出各目标物的子图像，对各目标物的子图像进行拼接，并将获取到的拼接后的图像作为待识别图像，如图2所示。
图2为本公开提供的确定待识别图像的示意图，图2中的左边部分为无人驾驶设备位于路口时采集到的图像，可见，该图像中有多个信号灯，于是，该无人驾驶设备可基于高精地图、车辆位姿、图像传感器位姿等，确定各信号灯的大致位置，即，图像中的区域A、B、C、D、E，并将区域A、B、C、D、E，作为各信号灯对应的第一区域。则该无人驾驶设备根据左边部分的区域A、B、C、D、E对应的图像，可确定出图2中右侧部分的待识别图像，可见，待识别图像由各目标物对应的第一区域的图像组成，区域A对应待识别图像的1部分，区域B对应待识别图像的2部分，区域C对应待识别图像的3部分，区域D对应待识别图像的4部分，区域E对待识别图对应的5部分。
需要说明的是，为了方便处理，待识别图像通常可设置为矩形，而不同的拼接规则可能会导致由各目标物的子图像拼接后的结果为非矩形，如图2中由1、2、3、4、5部分组成的多边形。因此，在本公开中，该无人驾驶设备还可针对各非矩形的拼接结果，将该图像进行补全。当然，具体的确定待识别图像的方法可根据需要进行设置，本公开对此不做限制。
S104：将所述待识别图像输入训练完成的目标检测模型，确定各目标物在所述待识别图像中分别对应的第二区域，以及各目标物的第一识别结果，其中，针对各所述目标物，该目标物对应的第二区域的范围小于该目标物对应的第一区域的范围。
在本公开提供的一个或多个实施例中，该无人驾驶设备在确定出待识别图像后，还可将该待识别图像输入训练完成的目标检测模型，确定各目标物在该待识别图像中的第二区域，以及各目标物的第一识别结果。
具体的，针对每个目标物，在步骤S100中确定出的第一区域为该目标物的大致范围。因此，基于各目标物分别对应的第一区域的图像，确定出的待识别图像，也限定了各目标物的大致范围。于是，在确定出待识别图像后，该无人驾驶设备可根据该待识别图像与预先训练好的目标检测模型，确定各目标物在该待识别图像中的具体位置，作为各目标物在该待识别图像中对应的第二区域，如图3所示。
图3为本公开提供的确定第二区域的示意图，图3中的左边部分为步骤S102中确定的待识别图像，中间部分为目标检测模型，右侧部分为该待识别图像中各目标物分别对应的第二区域，可见，各目标物分别对应的第二区域，为基于该待识别图像确定出的 各目标物的具体范围，因此，针对每个目标物，该目标物对应的第二区域的范围小于第一区域的范围。
在本公开提供的一个或多个实施例中，在确定各目标物分别对应的第二区域的同时，该无人驾驶设备还可通过目标检测模型，确定该待识别图像中的目标物的类型，作为各目标物的第一识别结果，如图4所示。
图4为本公开提供的确定第一识别结果的示意图，图4中的左边部分为步骤S102中确定的待识别图像，假设信号灯由上向下分别为黄灯、红灯、绿灯，白色代表熄灭，黑色代表亮灯，则各目标物的子图像对应的信号灯状态为绿色，图4中的中间部分为目标检测模型，右侧部分为该待识别图像中各目标物分别对应的第一识别结果，其中，第一识别结果为各目标物对应的信号灯状态为熄灭、信号灯状态为红色、信号灯状态为黄色、信号灯状态为绿色分别对应的概率。可见，各目标物对应的第一识别结果中，信号灯状态为绿色的概率较高。
S106：将各目标物分别对应的第二区域的图像输入训练完成的分类模型，确定各目标物的第二识别结果。
在本公开提供的一个或多个实施例中，由于步骤S104中确定出的各目标物的第一识别结果为根据待识别图像确定出的，而待识别图像中各目标物的子图像，限定了各目标物的大致范围。因此，可能出现由于待识别图像中的各目标物的子图像中非目标物的内容太多，导致第一识别结果不够准确的情况，因此，该无人驾驶设备，还可根据步骤S104中确定的各目标物对应的第二区域，即，各目标物的准确范围，确定各目标物分别对应的第二范围的图像，并将各目标物分别对应的第二区域的图像输入训练完成的分类模型，确定各目标物的第二识别结果。
具体的，针对每个目标物，该无人驾驶设备可根据步骤S104中确定出的该目标物对应的第二区域，从待识别图像中，提取该目标物对应的第二区域的图像，并将提取出的该目标物对应的第二区域的图像，输入到训练完成的分类模型中，确定该目标物的第二识别结果，如图5所示。
图5为本公开提供的确定第二识别结果的示意图，图5中左边部分为目标物对应的第二区域的图像，与图4类似，假设信号灯由上向下分别为黄灯、红灯、绿灯，白色代表熄灭，黑色代表亮灯，则目标物对应的第二区域的图像对应的信号灯状态为绿色，中间部分为分类模型，右边部分为第二识别结果，与第一识别结果类似，第二识别结果为 各目标物对应的信号灯状态为熄灭、信号灯状态为红色、信号灯状态为黄色、信号灯状态为绿色分别对应的概率。可见，目标物对应的第二区域的图像对应的第二识别结果中，信号灯状态为绿色的概率较高。
需要说明的是，本公开中的第一识别结果与第二识别结果，可指示各目标物是否为预设分类，如，信号灯状态是否为红色、信号灯状态是否为熄灭等，也可指示各目标物属于各预设分类的概率，如，信号灯状态为红色的概率、信号灯状态为绿色的概率等，具体的第一识别结果与第二识别结果的内容可根据需要进行设置，本公开对此不做限制。
S108：根据各目标物的第一识别结果以及第二识别结果，确定各目标物的最终识别结果。
在本公开提供的一个或多个实施例中，在确定出各目标物的第一识别结果以及第二识别结果后，该无人驾驶设备还可根据各目标物的第一识别结果以及第二识别结果，确定各目标物的最终识别结果。
具体的，在对目标物进行识别的过程中，步骤S104中确定的各目标物的第一识别结果，与步骤S106中确定的各目标物的第二识别结果，都可能出现识别不准确的情况，而若通过各目标物的第一识别结果与第二识别结果，以及各识别结果的权重，确定各目标物的最终识别结果，则可得到更为准确的识别结果。于是，可预设各目标物的第一识别结果的权重和第二识别结果的权重，并在需要对目标物进行识别时，针对每个目标物，根据该目标物的第一识别结果及其第一权重、以及第二识别结果及其第二权重，将该目标物的第一识别结果及其第一权重，以及第二识别结果及其第二权重，输入预先确定的加权确定函数
Figure PCTCN2022084308-appb-000001
其中，X 1为该目标物对应的第一识别结果，a 1为该第一识别结果对应的第一权重，X 2为该目标物对应的第二识别结果，a 2为该第二识别结果对应的第二权重，则可得到该目标物的加权结果，该加权结果为该目标物分别属于各预设分类的加权平均概率。
在确定出该目标物分别属于各预设分类的加权平均概率后，该无人驾驶设备可从确定出的该目标物分别属于各预设分类的加权平均概率中，确定概率最高的状态，作为该目标物的最终识别结果。如，信号灯对应的状态为熄灭、红色、黄色、绿色的加权平均概率分别为10％、80％、2％、8％，则可确定该目标物的最终识别结果对应的信号灯状态为红色。
基于图1的目标物识别方法，从采集的图像中确定各目标物的第一区域，根据各目 标物的第一区域的图像，确定待识别图像，通过目标检测模型，确定该待识别图像中各目标物的第二区域和第一识别结果，通过分类模型，从各目标物的第二区域的图像中，确定各目标物的第二识别结果，根据各目标物的第一识别结果以及第二识别结果，确定各目标物的最终识别结果。通过对采集图像中目标物对应区域的提取，减少噪声引起的识别准确率下降的问题，同时通过两阶段的识别，可以进一步更精准地确定每个目标物对应的区域，以及每个目标物的识别结果，最后通过两阶段的识别结果，确定最终识别结果，提高对目标物识别的准确率。
进一步的，在步骤S100中，根据高精地图确定出的目标物的位置仅为该目标物在图像中的大致位置，因此，为了保证确定出的第一区域的准确性，该无人驾驶设备还可根据放大参数值，分别对各目标物对应的区域进行放大，并将放大后的区域作为各目标物对应的第一区域。如图6所示。
图6为本公开提供的确定目标物对应的第一区域的示意图，根据高精地图可大致确定该目标物所在位置，实线框为以该目标物的原始尺寸确定出的该目标物的第一区域，可见，若根据以该目标物原始尺寸确定的该目标物的第一区域，不能准确确定出该目标物的识别结果。因此，该无人驾驶设备可根据放大参数值，对该目标物对应的区域进行放大，虚线框为根据放大参数值确定出的该目标物的第一区域。显然，根据放大后的第一区域的图像，在后续步骤中可准确确定该目标物的识别结果。其中，该目标物原始尺寸为根据无人驾驶设备当前位姿、高精地图等确定的，该放大参数值可为预设的。
更进一步地，在步骤S100中，根据放大参数值确定第一区域时，对于不同的目标物，放大参数值不完全相同，如，距离无人驾驶设备越近的目标物，若放大参数值过小，可能导致确定出的第一区域未完全包含该目标物，导致识别结果错误。而距离无人驾驶设备越远的目标物，若放大参数值过大，可能导致确定出的第一区域中非目标物的内容过多，导致识别结果错误。因此，该无人驾驶设备，在根据放大参数值对各目标物对应的区域进行放大时，还可针对每个目标物，根据该目标物的位置、该目标物的属性、以及采集图像时的采集设备的位姿，确定该目标物对应的放大参数值。
例如，根据该目标物的位置以及采集图像时的采集设备的位姿，确定该目标物与无人驾驶设备的距离，进而根据该距离，确定该目标物对应的放大参数值，其中，该距离与该放大参数值负相关。也就是说，该目标物与无人驾驶设备越近，该放大参数值越大，该目标物与无人驾驶设备越远，该放大参数值越小，以使距离较近的目标物对应的第一区域中完全包含该目标物，距离较远的目标物的第一区域包含的非目标物内容较少，进 而避免了由于待识别图像不准确导致识别结果错误的情况。或者，根据该目标物的属性，如具体为该目标物的尺寸，确定该目标物对应的放大参数值，其中，目标物的尺寸与放大参数值正相关。也就是说，同一图像中，按照相同的放大参数值确定第一区域时，可能出现尺寸较小的目标物对应的第一区域已完全包含该目标物、但尺寸较大的目标物对应的第一区域还未完全将该目标物包含在内的情况，因此，尺寸越大的目标物在确定第一区域时，其放大参数值越大。
另外，也可根据该目标物的属性中的其他信息，如，该目标物的形状，确定放大参数值，如，同一尺寸的圆形的目标物比三角形的目标物的放大参数值更小。当然，具体放大参数值的确定方式可根据需要进行设置，本公开对此不做限制。
需要说明的是，在无人驾驶设备行驶过程中，还可考虑如交通信息等情况，如，某路段的信号灯在当前时间段处于断电状态，则该无人驾驶设备在经过该路段时，可暂时不考虑信号灯对运动策略的影响。因此，各目标物的属性还可包括交通信息，具体的各目标物的属性还可根据需要进行设置，本公开对此不做限制，其中，交通信息可通过服务器下发或主动查询等方式获得。
进一步的，在确定出各目标物的第一区域和第二区域后，该无人驾驶设备还可根据各目标物的第一区域和第二区域，确定无人驾驶设备采集的图像中，各目标物所在位置并进行标注。于是，针对每个目标物，该无人驾驶设备可根据该目标物在待识别图像中对应的第二区域的位置，确定该目标物在对应的第一区域的图像中的位置。
然后，在确定出该目标物在第一区域的图像中的位置后，该无人驾驶设备还可根据该目标物在采集的图像中对应的第一区域的位置，以及该目标物在对应的第一区域的图像中的位置，确定目标物在所述采集的图像中的位置。
最后，在确定出该目标物在采集的图像中的位置后，该无人驾驶设备可根据该目标物在所述采集的图像中的位置，在采集的图像中标注该目标物。在采集的图像中标注该目标物可以是在采集的图像中标注该目标物的位置及识别结果，以便基于各目标物的位置及识别结果，确定无人驾驶设备的运动策略。
另外，在步骤S102中，该无人驾驶设备还可针对每个目标物的子图像，将该目标物的子图像，作为单独的待识别图像，并继续执行后续步骤，使得可分别对各待识别图像进行处理，分别确定每个目标物的最终识别结果。
上述为以无人驾驶设备为例对该说明书进行说明，在实际应用中，该目标物识别方 法还可由服务器执行，具体的，无人驾驶设备还可将该图像发送至检测目标物的服务器，由该服务器根据接收到的图像，对图像中的目标物进行识别，并将识别结果发送至无人驾驶设备或确定无人驾驶设备控制策略的服务器，确定该无人驾驶设备下一时刻的运动策略。
在本公开提供的一个或多个实施例中，本公开提供的目标物识别方法，可应用于确定无人驾驶设备运动策略时，确定无人驾驶设备的环境图像中的各目标物位置与类型的场景中，以使无人驾驶设备能够基于确定出的目标物的位置及类型，确定下一时刻该无人驾驶设备的运动策略，以使下一时刻无人驾驶设备正常行驶而不发生交通事故。具体的根据目标物类型与位置等确定无人驾驶设备运动策略的方法已经是现有较为成熟的技术，本公开对此不再赘述。
更进一步的，在本公开中，步骤S104中的目标检测模型，可由训练模型的服务器预先训练获得。在训练模型时，可获取通过由采集设备历史上采集到的若干图像，确定出的各目标物对应的第一区域确定的待识别图像，作为各训练样本，并对各训练样本中的各目标物进行标注，将各训练样本中的目标物的位置及类型作为训练标签。然后，该服务器可将各训练样本输入到待训练的目标检测模型中，确定各目标物的检测结果，以最小化各训练样本的检测结果与样本标签之间的差异为优化目标，对该目标检测模型进行训练。
另外，在本公开中，步骤S106中的分类模型，可由训练模型的服务器预先训练获得，在训练分类模型时，可获取若干有标注的目标物图像，作为训练样本。输入到待训练的分类模型中，确定各训练样本的分类结果，以最小化各训练样本的分类结果与标注之间的差异为优化目标，对该分类模型进行训练。
基于图1所示的目标物识别方法，本公开实施例还对应提供一种目标物识别装置的结构示意图，如图7所示。
图7为本公开实施例提供的目标物识别装置的结构示意图，所述装置包括：
确定模块200，用于根据预存的目标物的位置以及采集图像时的采集设备位姿，在采集的所述图像中分别确定各目标物对应的第一区域；
提取模块202，用于提取所述图像中各所述目标物对应的第一区域的图像，确定待识别图像；
第一识别模块204，用于将所述待识别图像输入训练完成的目标检测模型，确定各 所述目标物在所述待识别图像中的对应的第二区域，以及各所述目标物的第一识别结果，其中，针对各所述目标物，该目标物对应的第二区域范围小于该目标物对应的第一区域的范围；
第二识别模块206，用于将所述待识别图像中各所述目标物对应的第二区域的图像输入训练完成的分类模型，确定各所述目标物的第二识别结果；
识别模块208，用于根据各所述目标物的第一识别结果以及第二识别结果，确定各所述目标物的最终识别结果。
可选地，所述确定模块200，具体用于根据采集所述图像时的采集设备位姿，在预存的高精地图中，确定所述采集设备采集所述图像时，采集范围内的各所述目标物，根据各所述目标物的位置以及各所述目标物的属性，在采集的所述图像中分别确定各所述目标物对应的区域，根据放大参数值，分别对各所述目标物对应的区域进行放大，将放大后的区域作为各所述目标物对应的第一区域。
可选地，所述提取模块202，具体用于按照各所述目标物分别对应的第一区域，从所述图像中提取出各所述目标物的子图像，将对各所述目标物的子图像进行拼接后获得的图像，作为所述待识别图像。
可选地，所述第二识别模块206，具体用于针对每个所述目标物，根据所述目标检测模型输出的该目标物对应的第二区域，从所述待识别图像中，提取该目标物对应的第二区域的图像，将该目标物对应的第二区域的图像，输入训练完成的分类模型，确定该目标物的第二识别结果。
可选地，各所述目标物的第一识别结果以及第二识别结果为各所述目标物属于各预设分类的概率，所述识别模块208，具体用于针对每个所述目标物，根据该目标物的第一识别结果、第二识别结果、第一权重以及第二权重，确定该目标物分别属于各所述预设分类的加权平均概率，根据该目标物分别属于各所述预设分类的加权平均概率，确定该目标物的最终识别结果。
可选地，所述确定模块200，针对每个所述目标物，根据该目标物的位置、该目标物的属性以及采集图像时的所述采集设备位姿，确定该目标物对应的放大参数值，按照该目标物对应的放大参数值，对该目标物对应的区域进行放大。
可选地，所述识别模块208，具体用于针对每个所述目标物，根据该目标物在所述待识别图像中对应的第二区域的位置，确定该目标物在对应的第一区域的图像中的位置， 根据该目标物在所述采集的图像中对应的第一区域的位置，以及该目标物在对应的第一区域的图像中的位置，确定该目标物在所述采集的图像中的位置，根据该目标物在所述采集的图像中的位置，在所述采集的图像中标注该目标物。
本公开还提供了一种计算机可读存储介质，所述存储介质存储有计算机程序，所述计算机程序被处理器执行时可用于执行上述提供的目标物识别方法。
基于上述提供的目标物识别方法，本公开实施例还提供了图8所示的无人驾驶设备的结构示意图。如图8，在硬件层面，该无人驾驶设备包括处理器、内部总线、网络接口、内存以及非易失性存储器，当然还可能包括其他业务所需要的硬件。处理器从非易失性存储器中读取对应的计算机程序到内存中然后运行，以实现上述的目标物识别方法。
当然，除了软件实现方式之外，本公开并不排除其他实现方式，比如逻辑器件抑或软硬件结合的方式等等，也就是说以下处理流程的执行主体并不限定于各个逻辑单元，也可以是硬件或逻辑器件。
在20世纪90年代，对于一个技术的改进可以很明显地区分是硬件上的改进(例如，对二极管、晶体管、开关等电路结构的改进)还是软件上的改进(对于方法流程的改进)。然而，随着技术的发展，当今的很多方法流程的改进已经可以视为硬件电路结构的直接改进。设计人员几乎都通过将改进的方法流程编程到硬件电路中来得到相应的硬件电路结构。因此，不能说一个方法流程的改进就不能用硬件实体模块来实现。例如，可编程逻辑器件(Programmable Logic Device,PLD)(例如现场可编程门阵列(Field Programmable Gate Array，FPGA))就是这样一种集成电路，其逻辑功能由用户对器件编程来确定。由设计人员自行编程来把一个数字系统“集成”在一片PLD上，而不需要请芯片制造厂商来设计和制作专用的集成电路芯片。而且，如今，取代手工地制作集成电路芯片，这种编程也多半改用“逻辑编译器(logic compiler)”软件来实现，它与程序开发撰写时所用的软件编译器相类似，而要编译之前的原始代码也得用特定的编程语言来撰写，此称之为硬件描述语言(Hardware Description Language，HDL)，而HDL也并非仅有一种，而是有许多种，如ABEL(Advanced Boolean Expression Language)、AHDL(Altera Hardware Description Language)、Confluence、CUPL(Cornell University Programming Language)、HDCal、JHDL(Java Hardware Description Language)、Lava、Lola、MyHDL、PALASM、RHDL(Ruby Hardware Description Language)等，目前最普遍使用的是VHDL(Very-High-Speed Integrated Circuit Hardware Description Language)与Verilog。本领域技术人员也应该清楚，只需要将方法流程用上述几种硬件描述语言 稍作逻辑编程并编程到集成电路中，就可以很容易得到实现该逻辑方法流程的硬件电路。
控制器可以按任何适当的方式实现，例如，控制器可以采取例如微处理器或处理器以及存储可由该(微)处理器执行的计算机可读程序代码(例如软件或固件)的计算机可读介质、逻辑门、开关、专用集成电路(Application Specific Integrated Circuit，ASIC)、可编程逻辑控制器和嵌入微控制器的形式，控制器的例子包括但不限于以下微控制器：ARC 625D、Atmel AT91SAM、Microchip PIC18F26K20以及Silicone Labs C8051F320，存储器控制器还可以被实现为存储器的控制逻辑的一部分。本领域技术人员也知道，除了以纯计算机可读程序代码方式实现控制器以外，完全可以通过将方法步骤进行逻辑编程来使得控制器以逻辑门、开关、专用集成电路、可编程逻辑控制器和嵌入微控制器等的形式来实现相同功能。因此这种控制器可以被认为是一种硬件部件，而对其内包括的用于实现各种功能的装置也可以视为硬件部件内的结构。或者甚至，可以将用于实现各种功能的装置视为既可以是实现方法的软件模块又可以是硬件部件内的结构。
上述实施例阐明的系统、装置、模块或单元，具体可以由计算机芯片或实体实现，或者由具有某种功能的产品来实现。一种典型的实现设备为计算机。具体的，计算机例如可以为个人计算机、膝上型计算机、蜂窝电话、相机电话、智能电话、个人数字助理、媒体播放器、导航设备、电子邮件设备、游戏控制台、平板计算机、可穿戴设备或者这些设备中的任何设备的组合。
为了描述的方便，描述以上装置时以功能分为各种单元分别描述。当然，在实施本公开时可以把各单元的功能在同一个或多个软件和/或硬件中实现。
本领域内的技术人员应明白，本公开的实施例可提供为方法、系统、或计算机程序产品。因此，本公开可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本公开可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本公开是参照根据本公开实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中 指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
在一个典型的配置中，计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。
内存可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)。内存是计算机可读介质的示例。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
还需要说明的是，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、商品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、商品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、商品或者设备中还存在另外的相同要素。
本领域技术人员应明白，本公开的实施例可提供为方法、系统或计算机程序产品。 因此，本公开可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本公开可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本公开可以在由计算机执行的计算机可执行指令的一般上下文中描述，例如程序模块。一般地，程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、组件、数据结构等等。也可以在分布式计算环境中实践本公开，在这些分布式计算环境中，由通过通信网络而被连接的远程处理设备来执行任务。在分布式计算环境中，程序模块可以位于包括存储设备在内的本地和远程计算机存储介质中。
本公开中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于系统实施例而言，由于其基本相似于方法实施例，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
以上所述仅为本公开的实施例而已，并不用于限制本公开。对于本领域技术人员来说，本公开可以有各种更改和变化。凡在本公开的精神和原理之内所作的任何修改、等同替换、改进等，均应包含在本公开的权利要求范围之内。

Claims (10)

  1. 一种目标物识别方法，其特征在于，包括：
    从采集的图像中，确定各目标物分别对应的第一区域；
    根据所述图像中各所述目标物分别对应的第一区域的图像，确定待识别图像；
    将所述待识别图像输入训练完成的目标检测模型，确定各所述目标物在所述待识别图像中分别对应的第二区域，以及各所述目标物的第一识别结果，其中，针对各所述目标物，该目标物对应的第二区域的范围小于该目标物对应的第一区域的范围；
    将所述待识别图像中各所述目标物分别对应的第二区域的图像输入训练完成的分类模型，确定各所述目标物的第二识别结果；
    根据各所述目标物的第一识别结果以及第二识别结果，确定各所述目标物的最终识别结果。
  2. 如权利要求1所述的方法，其特征在于，从采集的所述图像中，分别确定各所述目标物对应的第一区域，包括：
    根据采集所述图像时的采集设备位姿，在预存的高精地图中，确定所述采集设备采集所述图像时，采集范围内的各所述目标物；
    根据各所述目标物的位置以及各所述目标物的属性，在采集的所述图像中分别确定各所述目标物对应的区域；
    根据放大参数值，分别对各所述目标物对应的区域进行放大，将放大后的区域作为各所述目标物对应的第一区域。
  3. 如权利要求1所述的方法，其特征在于，根据所述图像中各所述目标物分别对应的第一区域的图像，确定所述待识别图像，包括：
    按照各所述目标物分别对应的第一区域，从所述图像中提取出各所述目标物的子图像；
    将对各所述目标物的子图像进行拼接后获得的图像，作为所述待识别图像。
  4. 如权利要求1所述的方法，其特征在于，将所述待识别图像中各所述目标物分别对应的第二区域的图像输入训练完成的分类模型，确定各所述目标物的第二识别结果，包括：
    针对每个所述目标物，根据所述目标检测模型输出的该目标物对应的第二区域，从所述待识别图像中，提取该目标物对应的第二区域的图像；
    将该目标物对应的第二区域的图像，输入训练完成的分类模型，确定该目标物的第二识别结果。
  5. 如权利要求1所述的方法，其特征在于，各所述目标物的第一识别结果以及第二识别结果为各所述目标物属于各预设分类的概率；
    根据各所述目标物的第一识别结果以及第二识别结果，确定各所述目标物的最终识别结果，包括：
    针对每个所述目标物，根据该目标物的第一识别结果、第二识别结果、第一权重以及第二权重，确定该目标物分别属于各所述预设分类的加权平均概率；
    根据该目标物分别属于各所述预设分类的加权平均概率，确定该目标物的最终识别结果。
  6. 如权利要求2所述的方法，其特征在于，根据放大参数值，分别对各所述目标物对应的区域进行放大，包括：
    针对每个所述目标物，根据该目标物的位置、该目标物的属性以及采集图像时的所述采集设备位姿，确定该目标物对应的放大参数值；
    按照该目标物对应的放大参数值，对该目标物对应的区域进行放大。
  7. 如权利要求1所述的方法，其特征在于，所述方法还包括：
    针对每个所述目标物，根据该目标物在所述待识别图像中对应的第二区域的位置，确定该目标物在对应的第一区域的图像中的位置；
    根据该目标物在所述采集的图像中对应的第一区域的位置，以及该目标物在对应的第一区域的图像中的位置，确定该目标物在所述采集的图像中的位置；
    根据该目标物在所述采集的图像中的位置，在所述采集的图像中标注该目标物。
  8. 一种目标物识别装置，其特征在于，包括：
    确定模块，根据预存的目标物的位置以及采集图像时的采集设备位姿，在采集的所述图像中分别确定各目标物对应的第一区域；
    提取模块，提取所述图像中各所述目标物对应的第一区域的图像，确定待识别图像；
    第一识别模块，将所述待识别图像输入训练完成的目标检测模型，确定各所述目标物在所述待识别图像中的对应的第二区域，以及各所述目标物的第一识别结果，其中，针对各所述目标物，该目标物对应的第二区域范围小于该目标物对应的第一区域的范围；
    第二识别模块，将所述待识别图像中各所述目标物对应的第二区域的图像输入训练完成的分类模型，确定各所述目标物的第二识别结果；
    识别模块，根据各所述目标物的第一识别结果以及第二识别结果，确定各所述目标物的最终识别结果。
  9. 一种计算机可读存储介质，其特征在于，所述存储介质存储有计算机程序，所 述计算机程序被处理器执行时实现上述权利要求1-7任一项所述的方法。
  10. 一种无人驾驶设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其特征在于，所述处理器执行所述程序时实现上述权利要求1-7任一项所述的方法。
PCT/CN2022/084308 2021-04-07 2022-03-31 目标物识别 WO2022213867A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110369657.5 2021-04-07
CN202110369657.5A CN112766241B (zh) 2021-04-07 2021-04-07 一种目标物识别方法以及装置

Publications (1)

Publication Number Publication Date
WO2022213867A1 true WO2022213867A1 (zh) 2022-10-13

Family

ID=75691162

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/084308 WO2022213867A1 (zh) 2021-04-07 2022-03-31 目标物识别

Country Status (2)

Country Link
CN (1) CN112766241B (zh)
WO (1) WO2022213867A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115827880A (zh) * 2023-02-10 2023-03-21 之江实验室 一种基于情感分类的业务执行方法及装置
CN115880442A (zh) * 2023-02-06 2023-03-31 宝略科技(浙江)有限公司 一种基于激光扫描的三维模型重建方法以及系统

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112766241B (zh) * 2021-04-07 2021-07-30 北京三快在线科技有限公司 一种目标物识别方法以及装置
CN113344198B (zh) * 2021-06-09 2022-08-26 北京三快在线科技有限公司 一种模型训练的方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110287850A (zh) * 2019-06-20 2019-09-27 北京三快在线科技有限公司 一种模型训练以及目标物识别的方法及装置
CN110555511A (zh) * 2019-07-24 2019-12-10 北京踏歌智行科技有限公司 识别区域的方法、装置、电子设备和计算机可读存储介质
CN111797698A (zh) * 2020-06-10 2020-10-20 北京三快在线科技有限公司 目标物的识别方法及识别装置
CN112766241A (zh) * 2021-04-07 2021-05-07 北京三快在线科技有限公司 一种目标物识别方法以及装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107944351B (zh) * 2017-11-07 2020-08-04 深圳市易成自动驾驶技术有限公司 图像识别方法、装置及计算机可读存储介质
KR20200069084A (ko) * 2018-12-06 2020-06-16 팅크웨어(주) 도로 제한 속도 판단 방법, 도로 제한 속도 판단 장치, 전자 기기, 컴퓨터 프로그램 및 컴퓨터 판독 가능한 기록 매체
CN111563398A (zh) * 2019-02-13 2020-08-21 北京京东尚科信息技术有限公司 用于确定目标物的信息的方法和装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110287850A (zh) * 2019-06-20 2019-09-27 北京三快在线科技有限公司 一种模型训练以及目标物识别的方法及装置
CN110555511A (zh) * 2019-07-24 2019-12-10 北京踏歌智行科技有限公司 识别区域的方法、装置、电子设备和计算机可读存储介质
CN111797698A (zh) * 2020-06-10 2020-10-20 北京三快在线科技有限公司 目标物的识别方法及识别装置
CN112766241A (zh) * 2021-04-07 2021-05-07 北京三快在线科技有限公司 一种目标物识别方法以及装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115880442A (zh) * 2023-02-06 2023-03-31 宝略科技(浙江)有限公司 一种基于激光扫描的三维模型重建方法以及系统
CN115827880A (zh) * 2023-02-10 2023-03-21 之江实验室 一种基于情感分类的业务执行方法及装置

Also Published As

Publication number Publication date
CN112766241A (zh) 2021-05-07
CN112766241B (zh) 2021-07-30

Similar Documents

Publication Publication Date Title
WO2022213867A1 (zh) 目标物识别
WO2022213846A1 (zh) 识别模型的训练
WO2021052185A1 (zh) 确定智能驾驶车辆的行驶轨迹
CN111582189B (zh) 交通信号灯识别方法、装置、车载控制终端及机动车
CN111508258B (zh) 一种定位方法及装置
JP2021530048A (ja) 多階層化目標類別方法及び装置、交通標識検出方法及び装置、機器並びに媒体
WO2021218786A1 (zh) 一种数据处理系统、物体检测方法及其装置
US10762440B1 (en) Sensor fusion and deep learning
US11436815B2 (en) Method for limiting object detection area in a mobile system equipped with a rotation sensor or a position sensor with an image sensor, and apparatus for performing the same
JP2019220129A (ja) 交通標識を認識するためのシステムおよび方法
CN112327864A (zh) 一种无人驾驶设备的控制方法及控制装置
CN111797698A (zh) 目标物的识别方法及识别装置
CN114332808A (zh) 一种预测转向意图的方法及装置
CN114077252B (zh) 机器人碰撞障碍区分装置及方法
CN112990099B (zh) 一种车道线检测的方法以及装置
CN112699711A (zh) 车道线检测方法、装置、存储介质及电子设备
CN112818968A (zh) 一种目标物分类方法及装置
CN114972492A (zh) 一种基于鸟瞰图的位姿确定方法、设备和计算机存储介质
CN112861831A (zh) 一种目标物的识别方法、装置、存储介质及电子设备
CN116563376A (zh) 基于深度学习的lidar-imu紧耦合语义slam方法及相关装置
US20220340174A1 (en) Unmanned driving device control
CN114187355A (zh) 一种图像标定方法及装置
CN114895274A (zh) 一种护栏识别方法
CN114155504A (zh) 用于自动驾驶的视觉识别车辆方法、装置、出行装置及介质
CN113917453A (zh) 一种基于雷达和视频的多传感器融合方法

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: 22783934

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22783934

Country of ref document: EP

Kind code of ref document: A1