WO2018214706A1 - 运动控制方法、存储介质、计算机设备和服务机器人 - Google Patents

运动控制方法、存储介质、计算机设备和服务机器人 Download PDF

Info

Publication number
WO2018214706A1
WO2018214706A1 PCT/CN2018/085065 CN2018085065W WO2018214706A1 WO 2018214706 A1 WO2018214706 A1 WO 2018214706A1 CN 2018085065 W CN2018085065 W CN 2018085065W WO 2018214706 A1 WO2018214706 A1 WO 2018214706A1
Authority
WO
WIPO (PCT)
Prior art keywords
node
image
feature
image frame
map
Prior art date
Application number
PCT/CN2018/085065
Other languages
English (en)
French (fr)
Inventor
孟宾宾
Original Assignee
腾讯科技（深圳）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 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Publication of WO2018214706A1 publication Critical patent/WO2018214706A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation
    • G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
    • G05D1/02—Control of position or course in two dimensions
    • G05D1/021—Control of position or course in two dimensions specially adapted to land vehicles
    • G05D1/0231—Control of position or course in two dimensions specially adapted to land vehicles using optical position detecting means
    • G05D1/0246—Control of position or course in two dimensions specially adapted to land vehicles using optical position detecting means using a video camera in combination with image processing means
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
    • G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
    • G05D1/10—Simultaneous control of position or course in three dimensions
    • 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
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • 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
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation
    • G06V40/167—Detection; Localisation; Normalisation using comparisons between temporally consecutive images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification

Definitions

  • the present application relates to the field of computer technology, and in particular, to a motion control method, a storage medium, a computer device, and a service robot.
  • the motion control of the movable computer device is implemented based on the positioning of the sensor when the task is performed by a conventional mobile computer device.
  • the sensing signal is susceptible to the surrounding environment during the positioning process, which seriously affects the accuracy of the positioning, thereby causing the movement of the control computer device.
  • the accuracy rate is reduced.
  • a motion control method a storage medium, a computer device, and a service robot are provided.
  • a motion control method comprising:
  • the computer device acquires an image frame
  • the computer device When the computer device performs face detection on the image frame to obtain the image frame including a face image, determining that the face image is in a corresponding target node in the map;
  • the computer device picks a starting node that matches the image frame from the map; wherein a feature of the image frame matches a feature of a node image corresponding to the starting node;
  • the computer device selects a trending target motion path among the paths included in the map according to the start node and the target node;
  • the computer device moves in accordance with the selected target moving path.
  • One or more non-volatile storage media storing computer readable instructions, when executed by one or more processors, cause one or more processors to perform the following steps:
  • a computer device comprising a memory and a processor, the memory storing computer readable instructions, the computer readable instructions being executed by the processor such that the processor performs the following steps:
  • a service robot comprising a memory and a processor, the memory storing computer readable instructions, the computer readable instructions being executed by the processor, causing the processor to perform the following steps:
  • FIG. 1 is an application environment diagram of a motion control method in an embodiment
  • FIG. 2 is an internal structural diagram of a computer device for implementing a motion control method in an embodiment
  • FIG. 3 is a schematic flow chart of a motion control method in an embodiment
  • FIG. 5 is a schematic diagram of face recognition of a face image in an embodiment
  • FIG. 6 is a flow chart showing the steps of constructing a map in an embodiment
  • FIG. 7 is a schematic flow chart of a map creation process in an embodiment
  • Figure 8 is a schematic diagram of a map created in one embodiment
  • FIG. 9 is a schematic diagram of selecting a moving path of a target in a map in an embodiment
  • FIG. 10 is a schematic flow chart of a motion control method in another embodiment
  • Figure 11 is a block diagram showing the structure of a computer device in an embodiment
  • Figure 12 is a block diagram showing the structure of a computer device in another embodiment
  • Figure 13 is a block diagram showing the structure of a computer device in still another embodiment.
  • Figure 14 is a block diagram showing the structure of a computer device in still another embodiment.
  • FIG. 1 is an application environment diagram of a motion control method in an embodiment. As shown in FIG. 1, the motion control method is applied to a motion control system.
  • the motion control system is applied to indoor scenes.
  • the motion control system includes computer device 110 and target 120.
  • the computer device 110 can move to the target 120 by performing a motion control method.
  • the application environment shown in FIG. 1 is only a part of the scenario related to the solution of the present application, and does not constitute a limitation on the application environment of the solution of the present application.
  • the motion control system can also be applied to an outdoor open scene. .
  • the computer device includes a processor, a memory, a camera, a sound collection device, a speaker, a display screen, an input device, and a motion device connected by a system bus.
  • the memory comprises a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium of the computer device stores an operating system and can also store computer readable instructions that, when executed by the processor, cause the processor to implement a motion control method.
  • the processor is used to provide computing and control capabilities to support the operation of the entire computer device.
  • the internal memory can also store computer readable instructions that, when executed by the processor, cause the processor to perform a motion control method.
  • the display screen of the computer device may be a liquid crystal display or an electronic ink display screen.
  • the input device may be a touch layer covered on the display screen, or may be a button, a trackball or a touchpad provided on the terminal housing, or may be an external device. Keyboard, trackpad or mouse.
  • the computer device is a mobile electronic device, and may specifically be a service robot or the like. It will be understood by those skilled in the art that the structure shown in FIG. 2 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the terminal to which the solution of the present application is applied.
  • the specific terminal may include a ratio. More or fewer components are shown in the figures, or some components are combined, or have different component arrangements.
  • a motion control method is provided. This embodiment is mainly illustrated by the method being applied to the computer device in FIG. 2 described above. Referring to FIG. 3, the motion control method specifically includes the following steps:
  • the computer device can acquire an image frame in the current field of view of the camera through the camera to obtain the acquired image frame.
  • the field of view of the camera may vary due to changes in the posture and position of the computer device.
  • the computer device may acquire an image frame according to a fixed or dynamic frame rate to obtain an acquired image frame.
  • a fixed or dynamic frame rate enables a continuous dynamic picture to be formed when the image frame is played at the fixed or dynamic frame rate, so that the computer device can track a specific target in a continuous dynamic picture.
  • the computer device can invoke the camera to turn on the camera scan mode, scan a specific target in the current field of view in real time, and generate an image frame in real time according to a certain frame rate to obtain the generated image frame.
  • the computer device is a mobile electronic device, and specifically may be a robot or the like.
  • the camera can be a camera built into the computer device or an external camera associated with the computer device.
  • the camera can be a monocular camera, a binocular camera or an RGB-D (Red-Green-Blue-Deep) camera.
  • the face image is determined to be a corresponding target node in the map.
  • the map is a feature distribution map constructed by a computer device according to image frames acquired from natural space.
  • the computer device can construct a corresponding map for the natural space based on SLAM (Simultaneous Localization And Mapping and map construction).
  • the map constructed by the computer device based on SLAM may specifically be a three-dimensional point map.
  • a node is a location where a computer device projects a location from an image space in a natural space to a location in the map space.
  • the target node is the node where the target is projected to the map in a position in natural space.
  • the coordinates of the target in natural space are A(x1, y1, z1)
  • the coordinates after projecting A into the map space are B (x2, y2, z2), then B is the node of the target in the map.
  • the computer device may extract image data included in the image frame after acquiring the image frame, and detect whether the image data includes facial feature data. If the computer device detects that the face data is included in the image data, it is determined that the image frame includes a face image.
  • the computer device may also send the image frame to the server after acquiring the image frame, and the server completes the face detection process for the image frame, and then returns to the computer device whether the image frame includes the detection result of the face image.
  • the detection result may include a probability that a face image exists in the image frame and a coordinate area of the face image.
  • a number of nodes may be included in the map, and each node has a one-to-one corresponding node image.
  • the map may also include feature points extracted from the node image.
  • a map including feature points and nodes is a three-dimensional reconstruction of a scene in natural space. Specifically, the three-dimensional points in the three-dimensional scene in the natural space are transformed by the projection matrix to obtain pixel points in the two-dimensional image frame of the camera plane of the computer device, and the pixels in the two-dimensional image frame are projected through the projection matrix. The inverse transform obtains the three-dimensional feature points in the three-dimensional reconstruction scene in the map.
  • the computer device may calculate the position of the face image in the map when detecting that the face image is included in the image frame. Specifically, the computer device may determine a coordinate position of the face image in the image frame, calculate a position of the face image in the map according to a projection matrix adapted to a camera of the computer device, and search for a node included in the map. The node corresponding to the calculated position gets the target node.
  • the computer device may extract a background feature point of the background image in the image frame when the image of the image is included in the image frame, and match the extracted background feature point with the feature point included in the map. The position of the feature point in the map that matches the extracted background feature point is obtained, so that the node closest to the location is selected in the map to obtain the target node.
  • the image frame acquired by the computer device may be two or more frames of image.
  • the computer device detects the presence of the face image in the acquired image, the similarity matrix between the image frames of any two frames can be calculated, and the matching face is selected from the face image included in the image frame for calculating the similarity matrix.
  • the computer device can further determine the similarity matrix between any two frames of image frames and the position of the selected facial feature points on the two frames of images, and determine the facial feature points in the natural space according to the triangulation algorithm. position.
  • the computer device can further determine the position of the face feature point in the map according to the position of the face feature point in the natural space, thereby selecting a node closest to the location in the map to obtain the target node.
  • the node image is an image acquired by the computer device at a position in a natural space having a projection relationship with a node in the map.
  • the features of the image may be one or a combination of color features, texture features, and shape features.
  • the computer device may extract features of the node images corresponding to the nodes in the map when the map is constructed, and store the features of the extracted node images in a database or a cache with respect to the corresponding nodes.
  • the computer device can generate a one-dimensional image feature vector characterizing the features of the acquired image frame in a manner that characterizes the image of the node.
  • the computer device may recalculate the vector similarity between the generated one-dimensional image feature vector and the one-dimensional image feature vector corresponding to each node of the map, and further determine whether the vector similarity is greater than or equal to the preset vector similarity; if yes, match If not, it does not match.
  • a path formed by nodes in the map may be included in the map.
  • the computer device can start the node as the starting point and the target node as the end point, and select the path in the path formed by the node in the map to obtain the moving path toward the target.
  • the motion control method automatically determines the target node corresponding to the face image in the map when the image frame includes the face image, and locates the position of the target in the map, and then Based on the matching relationship between the feature of the image frame and the feature of the node image corresponding to each node in the map, the starting node matching the image frame can be selected from the map, and the current position of the local device in the map is located, and then According to the current node and the target node, the trending target moving path can be selected to move in the path included in the map. In this way, the positioning in the map can be completed by the feature matching between the images, the environmental influence caused by the positioning of the sensing signal is avoided, and the accuracy of the motion control is improved.
  • the convolutional neural network model is a complex network model formed by multiple layers connected to each other.
  • the neural network model may include a multi-layer feature conversion layer.
  • Each layer of the feature conversion layer has a corresponding nonlinear variation operator.
  • the nonlinear variation operator of each layer may be multiple, and one nonlinear variation in each layer of the feature conversion layer
  • the child makes a nonlinear change to the input image, and obtains a feature map as a result of the operation.
  • the convolutional neural network model is a model for extracting facial features obtained by performing learning training using an image including a face image as training data.
  • the computer device After acquiring the image frame, the computer device inputs the image frame into the convolutional neural network model, and uses the convolutional neural network model to extract the facial features of the image frame.
  • the face feature may be one or more features for reflecting the gender of the person, the outline of the face, the hairstyle, the glasses, the nose, the mouth, and the distance between the various facial organs.
  • the convolutional neural network model is a model for extracting image features obtained by learning training using images as training data. After acquiring the image frame, the computer device inputs the image frame into the convolutional neural network model, and uses the convolutional neural network model to perform image feature extraction on the image frame.
  • the computer device may acquire a feature map of the plurality of network layer outputs included in the convolutional neural network model.
  • the feature map is composed of response values obtained by processing the input image by a nonlinear variation operator. Different network layers extract different features.
  • the computer device can determine the corresponding facial feature data of the input image by using the feature map output by the convolutional neural network that extracts the facial features.
  • the computer device can determine the image feature data corresponding to the input image by using the feature map output by the convolutional neural network that extracts the image feature, and further determine whether the face feature data is included in the image feature data.
  • the computer device may perform image processing by using a 52-layer depth residual network model, and extract a feature map of the 4-layer fully connected layer output included in the depth residual network model as a subsequent input.
  • the memory neural network model is a neural network model that can comprehensively process sequence input.
  • the memory neural network model is a recurrent neural network model.
  • the memory neural network model may specifically be LSTM (Long Short-Term Memory).
  • the computer device can sequentially input the acquired feature maps into the memory neural network model to perform face feature detection.
  • the computer device can obtain the face detection result obtained by the memory neural network model according to the input feature map synthesis processing.
  • the face detection result includes the probability of the presence of the face image and the coordinate area of the face image in the image frame.
  • the computer device may also filter out the face of the overlapping area exceeding the preset overlap threshold according to the coordinate area of the face image included in the face detection result in the image frame.
  • the detection result is obtained according to the face detection result retained after filtering to obtain the coordinate area of the face image in the image frame.
  • the memory neural network model may use a rectangular window to move in the input feature map according to a preset direction and a preset step size, thereby performing window scanning, and extracting facial features in the scanned window image during scanning.
  • the data according to the extracted face feature image, obtains the probability that a face image exists in the scanned window image.
  • the window image in which the calculated probability is ranked first is stored in the coordinate area in the image frame, and the subsequent input feature map is processed.
  • FIG. 5 is a diagram showing face recognition of a face image in one embodiment.
  • the memory neural network model adopted by the computer device scans and analyzes the input feature map according to the rectangular window, and obtains the probability P A of the face image corresponding to the rectangular window A, and the probability of the presence of the face image corresponding to the rectangular window B.
  • P B the probability P C of the presence of the face image corresponding to the rectangular window C.
  • the memory neural network model can record the rectangular window C corresponding to the P C , continue to scan the feature map of the subsequent input according to the rectangular window, and comprehensively analyze the rectangular window and Corresponding to the probability of the presence of the face image, the probability of the presence of the face image in the image frame acquired by the computer device and the coordinate region of the face image in the image frame are output.
  • the image features are fully extracted by the plurality of network layers included in the convolutional neural network model, and the feature input from the multi-layer network layer is integrated into the memory neural network model to make the face detection more accurate.
  • the motion control method further includes a step of recognizing a face, the step of recognizing the face includes: extracting facial feature data of the facial image; and querying the pre-matching of the facial image according to the facial feature data Setting a face image; obtaining a target identification result according to the preset face image; and determining a service type associated with the target identification result.
  • the step of the face recognition described above may be specifically performed after S304.
  • the motion control method further includes: providing a service trigger entry corresponding to the type of service. This step can be specifically performed after S310.
  • the target identification result is data used to reflect the target identity.
  • the target identity can be the name of the target, social status, or job information.
  • a preset face image library is disposed on the computer device, and the preset face image library includes a plurality of preset face images.
  • the computer device may compare the face image in the image frame with the preset face image included in the preset face image library when detecting the face image in the image frame, and detect the face image and the pre-image in the image frame. Set whether the face images match.
  • the computer device matches between the face image and the preset face image in the image frame, it is determined that the face image included in the image frame and the preset face image are the same person image, and the preset face image is acquired.
  • the corresponding target identity information is used as the target identification result.
  • the preset face image may be a real face image for reflecting the corresponding target.
  • the selected image may be customized by the corresponding target from the personal data uploaded by the target or the historically published image information, or the selected image may be automatically analyzed by the system as the corresponding preset facial image.
  • the computer device detects whether the face image in the image frame matches the preset face image, and specifically calculates the similarity between the face image and the preset face image in the image frame.
  • the computer device may first extract the respective features of the face image and the preset face image in the image frame, thereby calculating the difference between the two features. The greater the difference between the features, the lower the similarity, and the difference between the features. Smaller, the higher the similarity.
  • an acceleration algorithm suitable for the image processor may be used to increase the operation rate.
  • the computer device may extract facial feature data from the image data after determining that the image is included in the image frame, and then extract the extracted facial feature data and each preset in the preset facial image database. The face feature data corresponding to the face image is compared to obtain the target identification result.
  • the image frame obtained by the computer device detecting the image frame may include one or more face images.
  • the computer device may determine a proportion of the face image included in the image frame to the image frame, extract face feature data of the face image that exceeds a preset ratio; and/or determine a face image included in the image frame. Sharpness, extracting face feature data of a face image whose resolution exceeds the sharpness threshold. The computer device then recognizes the face image from which the face feature data is extracted.
  • the computer device can find the service type associated with the target identification result.
  • the service type is the type to which the service provided to the target belongs. Service types such as restaurant order service or hotel reception service.
  • the service type can be a type that is uniformly set, a type related to the target identity, or a type related to the target attribute.
  • the computer device can set the service type in advance, associate the service type with the target identification, store the set service type in a database or file, and read from a database or file as needed. After identifying the target identification result, the computer device may pull the service type associated with the target identifier corresponding to the target identification result.
  • the computer device may provide the service trigger entry corresponding to the determined service type to the target after moving to the target.
  • the computer device can provide a service trigger entry through the display screen, and can also provide a voice service portal with the target through the speaker and sound collector.
  • the computer device may acquire an image frame to determine the current location, and provide a service trigger entry for the target, and receive the input service parameter through the display screen or the sound collector, and the computer device determines The object of the current service, the location of the current service, and the content of the current service.
  • the service portal associated with the target can be provided to the target. , greatly improving the efficiency of service provision.
  • the face recognition step processed by the computer device and the service type associated with determining the target identification result may be processed by the server.
  • the computer device may send the acquired image frame to the server, and after completing the face detection, face recognition, and determining the service type associated with the target identification result, the server identifies the target identification result and the associated service type. Send to computer device.
  • the motion control method further includes the step of constructing a map. It can be understood that the step of constructing the map can be performed before S402.
  • the steps of constructing the map specifically include:
  • S602. Select an image frame from the image frames acquired in time series.
  • the selected image frame may be a key frame in the captured image frame.
  • the computer device can receive a user selection instruction to select an image frame from the acquired image frames in accordance with the user selection instruction.
  • the computer device may select an image frame from the acquired image frames in accordance with a preset number of spaced frames. For example, an image frame is selected every 20 frames of image frames.
  • the feature of the preset node image is a preset feature for selecting a node image.
  • the feature of the node image that conforms to the preset may be that the number of feature points included in the feature points included in the image and the feature points included in the image of the existing node exceeds a preset number, or may be a feature point included and an image of an existing node.
  • the proportion of the matched feature points in the included feature points to the feature points included in the existing node image is lower than the preset ratio.
  • the recently added node image includes 100 feature points
  • the currently selected image frame includes 120 feature points.
  • the preset number is 50 and the preset ratio is 90%.
  • the number of feature points included in the feature point included in the currently selected image frame and the feature point included in the recently added node image is 70.
  • the number of feature points included in the current image frame and the feature points included in the existing node image exceeds a preset number, and it may be determined that the feature of the currently selected image frame conforms to the feature of the preset node image.
  • the selected image frame is obtained as a node image.
  • the computer device may acquire the image frame according to a fixed or dynamic frame rate, and select an image frame that is included in the captured image frame to have an initial number of feature points greater than a preset number threshold.
  • the node image determines a corresponding node of the node image in the map, and a corresponding position of the feature point included in the node image in the map, and constructs a partial map.
  • the computer device selects an image frame from the image frames acquired in time series, and selects an image frame that conforms to the feature of the preset node image as a subsequent node image until a global map is obtained.
  • the computer device may use the initial node image as a reference node image to track feature points in the reference node image.
  • the selected image frame includes a feature point that is less than the first preset number and is higher than the second preset number
  • the selected image frame is used as the node image.
  • the recently acquired node image is used as the reference node image, and image tracking is continued to select the node image.
  • the computer device may determine to acquire a node in the natural space where the acquired node image is projected in the map space.
  • the computer device can extract the feature of the node image in front of the acquired node image timing, calculate the feature of the node image with the highest timing and the change matrix of the acquired node image, and obtain the node image of the front-end acquisition timing according to the change matrix
  • the position is changed to the position when the acquired node image is acquired, and the acquired node image is determined to be a corresponding node in the map according to the change amount.
  • step S608 includes: extracting features of the acquired node image; acquiring features of the node image corresponding to the existing nodes in the map; determining a change matrix between the acquired features and the extracted features; A matrix that determines the node of the acquired node image in the map.
  • the change matrix is a similar change relationship between the features of the two-dimensional image and the features of the two-dimensional image.
  • the computer device may extract features of the acquired node image, and match features of the node images corresponding to the existing nodes in the map, and obtain locations of the successfully matched features in the acquired node image and the existing node image.
  • the acquired node image is an image frame acquired later
  • the existing node image is an image frame acquired later.
  • the computer device can determine the change matrix between the two image frames collected successively according to the obtained matching features on the positions of the two image frames collected successively, thereby obtaining the position change of the computer device when acquiring the two frame image frames and The posture changes, and then according to the position and posture of the previously acquired image, the position and posture of the image acquired later can be obtained.
  • the node image corresponding to the existing node in the map may be one frame or multiple frames.
  • the computer device can also compare the acquired feature of the node image with the feature of the node image corresponding to the plurality of existing nodes to obtain a change matrix of the captured image frame and the plurality of previously acquired image frames, and then according to the plurality of The change matrix is integrated to obtain the position and pose of the image acquired later. For example, the calculated plurality of position changes and posture changes are weighted and averaged.
  • the transformation relationship between the currently acquired node image and the previously existing node image is obtained by the change matrix between the features of the node image, thereby realizing the current position estimation by the position of the previous image frame in the map.
  • the position of the image frame in the map enables real-time positioning.
  • the feature of the acquired node image is stored corresponding to the determined node.
  • the computer device may extract features of the node image, and store the feature of the node image corresponding to the node corresponding to the node image, and may directly search for the feature of the corresponding node image according to the node when the image feature comparison is needed to save storage space. Improve search efficiency.
  • the map construction can be automatically performed, thereby avoiding the need for a large number of workers with professional drawing capabilities to manually survey the environment, and has high requirements on staff capabilities. And the problem of large amount of labor, improve the efficiency of map construction.
  • the motion control method further includes: calculating a similarity between a feature of the node image corresponding to the existing node in the map, and a feature of the acquired node image; a node corresponding to the existing node in the local map
  • the similarity between the feature of the image and the acquired feature of the node image exceeds the preset similarity threshold, the circular path including the existing node is generated in the map according to the node corresponding to the acquired node image. It can be understood that these steps can be specifically performed after S608.
  • the computer device when the computer device acquires the node image, the newly added feature of the node image is compared with the feature of the node image corresponding to the existing node in the map, and the feature of the newly added node image and the existing image in the map are calculated.
  • the similarity between the features of the node image corresponding to the node When the similarity between the feature of the node image corresponding to the existing node in the map and the feature of the newly added node image exceeds the preset similarity threshold, the computer device may determine that the newly added node image is collected in the natural space.
  • the node image corresponding to the existing node has the same collection position in the natural space.
  • the computer device may generate, by the corresponding node of the acquired node image, a node added from the existing node in the map, after the existing node, and a ring path from the existing node.
  • the computer device may sequentially acquire the features of the node images corresponding to the nodes included in the ring path in sequence from the existing nodes; and sequentially determine the change matrix between the acquired features of the node images of the corresponding adjacent nodes; The change matrix adjusts the features of the node image corresponding to each node included in the ring path in reverse order.
  • the computer device sequentially adds a node image to construct a partial map from the first frame node image.
  • determining a collection position of the fourth frame node image in the natural space and the first frame node The acquisition position of the image in the natural space is consistent, and a circular path of the first frame node image - the second frame node image - the third frame node image - the first frame node image is generated.
  • the change matrix between the feature of the first frame node image and the feature of the second frame node image is H1
  • the change matrix between the feature of the second frame node image and the feature of the third frame node image is H2
  • the third The variation matrix between the feature of the frame node image and the feature of the fourth frame node image is H4.
  • the computer device may change the feature of the first frame node image according to H4, optimize the third frame node image according to the feature of the obtained image, and then change the optimized third frame node image according to H3, according to the feature optimization of the obtained image. Two-frame node image.
  • the closed-loop detection is performed based on the similarity between the feature of the newly added node image and the feature of the existing node image.
  • a closed loop is detected, a circular path is generated in the map for subsequent closed loop. Optimize and improve the accuracy of building maps.
  • Figure 7 shows a flow diagram of a map creation process in one embodiment.
  • the map creation process includes three parts: tracking, mapping, and closed loop detection.
  • the computer device can acquire the image frame according to a fixed or dynamic frame rate.
  • the feature points of the image frame are extracted, and the extracted feature points are matched with the feature points of the node image corresponding to the newly added nodes in the map.
  • the computer device may reacquire the acquired image frames for relocation.
  • the image frames collected according to the newly added nodes in the map are corresponding to the nodes in the map.
  • the computer device can re-track the feature points in the map that match the acquired image, and optimize the image frame corresponding to the nodes in the map according to the matched features. After the optimized image is completed, it is determined whether the feature point of the image frame meets the feature point of the preset node image. If not, the computer device may reacquire the acquired image frame for feature point matching.
  • the computer device may acquire the image frame as a new node image.
  • the computer device may extract the feature points of the newly added node image, represent the extracted feature points according to a preset unified format, and then determine the position of the feature points of the newly added node image in the map according to the triangulation algorithm, thereby updating
  • the local map is further adjusted by local bundling to remove redundant nodes corresponding to the node image whose similarity is higher than the preset similarity threshold.
  • the closed loop detection can be performed asynchronously.
  • the feature of the newly added node image is compared with the feature of the node image corresponding to the existing node, and the similarity between the feature of the newly added node image and the feature of the node image corresponding to the existing node is higher than the preset.
  • the similarity threshold the computer device can determine that the collection position of the newly added node image in the natural space is consistent with the collection position of the node image corresponding to the existing node in the natural space, that is, there is a closed loop.
  • the computer device can further generate a circular path including nodes with the same position according to the corresponding node of the newly added node image, and perform closed-loop optimization and closed-loop fusion. Finally get a global map including feature points, nodes and paths
  • Figure 8 shows a schematic diagram of a map created in one embodiment.
  • the map is a schematic diagram of feature distribution based on sparse features.
  • the schematic diagram includes feature points 801, nodes 802, and paths 803 formed between the nodes.
  • the feature point 801 is a projection position of the position of the feature point of the object in the natural space in the natural space in the map space.
  • Node 802 is the projected position of the natural spatial location of the computer device in the map space when the image frame is acquired in natural space.
  • the path 803 formed between the nodes is a projection of the path of the computer device moving in the natural space in the map space.
  • step S306 includes: extracting features of the image frame; acquiring features of the node image corresponding to the node included in the map; determining a similarity between the feature of the image frame and the feature of the node image; and selecting the highest similarity The node corresponding to the feature of the node image obtains the starting node that matches the image frame.
  • the computer device compares the feature of the node image corresponding to the node existing in the map with the feature of the acquired node image, the difference between the two image features may be calculated, and the difference between the features is similar.
  • the similarity may be a cosine similarity or a Hamming distance in which the respective perceived hash values are between the images.
  • the current position of the node image corresponding to the node included in the map is similarly matched to locate the current position in the map, so that the positioning result is more accurate.
  • the motion control method further includes: extracting features of the image frame; acquiring features of the node image corresponding to the start node; determining a spatial state difference between the feature of the image frame and the feature of the node image; The amount of spatial state difference is exercised. It will be appreciated that these steps can be performed prior to S310.
  • the spatial state difference is the amount of change in the spatial state of the computer device when acquiring different image frames.
  • the spatial state difference amount includes the spatial position difference amount and the spatial angle difference amount.
  • the amount of spatial position difference is the movement of the computer device at a physical location. For example, the computer device horizontally shifts forward by 0.5 m when acquiring the first frame of image frames until the second frame of image frames is acquired.
  • the spatial angle difference is the rotation of the computer device in a physical orientation. For example, the computer device rotates 15 degrees counterclockwise when acquiring the first frame of the image frame until the second frame of the image frame is acquired.
  • the computer device may calculate a change matrix between the feature of the image frame and the feature of the node image corresponding to the start node, recover the motion position of the computer device according to the calculated change matrix, and decompose the rotation matrix from the change matrix.
  • the displacement matrix obtains the spatial angular difference between the feature of the image frame and the feature of the node image according to the rotation matrix, and obtains the spatial position difference between the feature of the image frame and the feature of the node image according to the displacement matrix.
  • the computer device can further determine the direction of the current motion according to the spatial angle difference amount, determine the distance of the current motion according to the spatial position difference amount, and thereby move the determined distance according to the determined direction.
  • the amount of spatial state difference between the currently acquired image frame and the node image corresponding to the determined initial node is moved to the initial node in the map, thereby moving to the target according to the selected trending target moving path. Exercise ensures the accuracy of the exercise.
  • the step S310 includes: sequentially acquiring features of the node image corresponding to the nodes included in the target moving path; and sequentially determining the spatial state difference between the acquired features of the node images of the corresponding neighboring nodes; The amount of spatial state difference determined in turn is exercised.
  • the computer device may acquire a feature of the node image corresponding to the second node that is adjacent to the initial node that is included in the target motion path, and calculate a feature of the node image corresponding to the initial node and a feature of the node image corresponding to the second node.
  • the computer equipment further decomposes the change matrix to obtain a rotation matrix and a displacement matrix. According to the rotation matrix, the spatial angle difference between the feature of the image frame and the feature of the node image is obtained, and the feature of the image frame and the feature of the node image are obtained according to the displacement matrix. The amount of space difference between the locations.
  • the computer device can further determine the direction of the current motion according to the spatial angle difference amount, determine the distance of the current motion according to the spatial position difference amount, thereby moving the determined distance according to the determined direction, and moving to the second node in the map.
  • the computer device can further determine the distance and direction of the current motion according to the same processing manner, and sequentially move from the second node in the map to the moving path toward the target until reaching the target node.
  • the spatial state difference amount of the feature of the node image corresponding to the adjacent node included in the target motion path is gradually moved from the starting node to the target node on the map according to the trending target moving path, thereby avoiding Deviations during the movement cannot determine the current position and ensure the accuracy of the movement.
  • Figure 9 is a diagram showing the selection of a trending target motion path in a map in one embodiment.
  • the schematic diagram includes a target node 901, a start node 902, and a trend target motion path 903.
  • the computer device After determining the location of the target node 901, that is, the location of the target node and the location of the local node 902, the computer device takes the start node 902 as the starting point and the target node 901 as the end point, and selects the target motion path 903 in the map.
  • the motion control method includes the following steps:
  • S1002 Select an image frame from image frames acquired in time series.
  • step S1004 Determine whether the feature of the selected image frame meets the feature of the preset node image; if yes, go to step S1006; if no, return to step S1002.
  • S1006 Acquire the selected image frame as a node image.
  • S1008 extracting features of the acquired node image; acquiring features of node images corresponding to existing nodes in the map; determining a change matrix between the acquired features and the extracted features; determining, according to the node and the change matrix, the acquired node image Corresponding nodes in the map correspond to the characteristics of the node image obtained by the determined node storage.
  • S1010 calculating a similarity between a feature of the node image corresponding to the existing node in the map and a feature of the acquired node image; a feature of the node image corresponding to the existing node in the local map, and a feature of the acquired node image
  • a circular path including the existing node is generated in the map according to the node corresponding to the acquired node image.
  • S1012 Acquire an image frame.
  • step S1016 Determine whether the face detection result indicates that the image frame includes a face image; if yes, go to step S1018; if no, return to step S1012.
  • S1020 Determine a corresponding target node of the face image in the map.
  • S1028 acquiring, in sequence, a feature of a node image corresponding to each node included in the target motion path; determining, in sequence, a spatial state difference between the acquired feature of the node image of the corresponding adjacent node; and performing, according to the spatial state difference amount determined in sequence motion.
  • the target node corresponding to the face image is determined in the map, and the position of the target in the map is determined. Then, based on the matching relationship between the feature of the image frame and the feature of the node image corresponding to each node in the map, the starting node matching the image frame can be selected from the map, and the current position of the local device in the map is located. According to the current node and the target node, the trending target moving path can be selected to move in the path included in the map. In this way, the positioning in the map can be completed by the feature matching between the images, the environmental influence caused by the positioning of the sensing signal is avoided, and the accuracy of the motion control is improved.
  • the various steps in the various embodiments of the present application are not necessarily performed in the order indicated by the steps. Except as explicitly stated herein, the execution of these steps is not strictly limited, and the steps may be performed in other orders. Moreover, at least some of the steps in the embodiments may include multiple sub-steps or multiple stages, which are not necessarily performed at the same time, but may be executed at different times, and the execution of these sub-steps or stages The order is also not necessarily sequential, but may be performed alternately or alternately with other steps or at least a portion of the sub-steps or stages of the other steps.
  • a computer device 1100 is provided.
  • the internal structure of the computer device 1100 can refer to the structure shown in FIG. 2.
  • Each of the modules described below can be implemented in whole or in part by software, hardware, or a combination thereof.
  • the computer device 1100 includes an acquisition module 1101, a determination module 1102, a selection module 1103, a selection module 1104, and a motion module 1105.
  • the obtaining module 1101 is configured to acquire an image frame.
  • the determining module 1102 is configured to determine, when the image frame is subjected to face detection, that the image frame includes a face image, determine a corresponding target node of the face image in the map.
  • the selection module 1103 is configured to select a starting node that matches the image frame from the map; wherein the feature of the image frame matches the feature of the node image corresponding to the starting node.
  • the selecting module 1104 is configured to select a trending target moving path among the paths included in the map according to the starting node and the target node.
  • the motion module 1105 is configured to move according to the selected trend target motion path.
  • the computer device 1100 after acquiring the image frame, automatically determines the target node corresponding to the face image in the map when the image frame includes the face image is detected, and locates the position of the target in the map, and then Based on the matching relationship between the feature of the image frame and the feature of the node image corresponding to each node in the map, the starting node matching the image frame can be selected from the map, and the current position of the local device in the map is located, and then According to the current node and the target node, the trending target moving path can be selected to move in the path included in the map. In this way, the positioning in the map can be completed by the feature matching between the images, the environmental influence caused by the positioning of the sensing signal is avoided, and the accuracy of the motion control is improved.
  • computer device 1100 further includes a detection module 1106.
  • the detecting module 1106 is configured to input the image frame into the convolutional neural network model; obtain the feature maps of the plurality of network layer outputs included in the convolutional neural network model; sequentially input the feature maps into the memory neural network model; and obtain the memory neural network model output Whether the image frame includes the result of the face image.
  • the image features are fully extracted by the plurality of network layers included in the convolutional neural network model, and the feature input from the multi-layer network layer is integrated into the memory neural network model to make the face detection more accurate.
  • computer device 1100 further includes an identification module 1107 and a service module 1108.
  • the recognition module 1107 is configured to extract face feature data of the face image, and query the preset face image that matches the face image according to the face feature data; obtain the target identity recognition result according to the preset face image; determine and target The type of service associated with the identity result.
  • the service module 1108 is configured to provide a service trigger entry corresponding to the service type.
  • the existing face is identified, and after identifying the identity of the target and moving to the target, the target can be provided with the service associated with the target.
  • the entrance greatly improves the efficiency of service provision.
  • computer device 1100 further includes a map construction module 1109.
  • the map construction module 1109 is configured to: select an image frame from the image frames acquired in time series; determine whether the feature of the selected image frame conforms to a feature of the preset node image; and when the feature of the selected image frame conforms to the feature of the node image, Obtaining the selected image frame as a node image; determining the acquired node image in a corresponding node in the map; and corresponding to the determined node storing the acquired feature of the node image.
  • the map construction can be automatically performed, thereby avoiding the need for a large number of workers with professional drawing capabilities to manually survey the environment, and has high requirements on staff capabilities. And the problem of large amount of labor, improve the efficiency of map construction.
  • the map construction module 1109 is further configured to extract features of the acquired node image; acquire features of the node image corresponding to the existing nodes in the map; and determine a change matrix between the acquired features and the extracted features; The node and the change matrix determine the node of the acquired node image in the map.
  • the transformation relationship between the currently acquired node image and the previously existing node image is obtained by the change matrix between the features of the node image, thereby realizing the current position estimation by the position of the previous image frame in the map.
  • the position of the image frame in the map enables real-time positioning.
  • the map construction module 1109 is further configured to calculate a similarity between a feature of the node image corresponding to the existing node in the map and a feature of the acquired node image; a node corresponding to the existing node in the local map When the similarity between the feature of the image and the acquired feature of the node image exceeds the preset similarity threshold, the circular path including the existing node is generated in the map according to the node corresponding to the acquired node image.
  • the closed-loop detection is performed based on the similarity between the feature of the newly added node image and the feature of the existing node image.
  • a closed loop is detected, a circular path is generated in the map for subsequent closed loop. Optimize and improve the accuracy of building maps.
  • the selection module 1103 is further configured to extract features of the image frame; acquire features of the node image corresponding to the node included in the map; determine a similarity between the feature of the image frame and the feature of the node image; The node corresponding to the feature of the node image with the highest degree obtains the starting node that matches the image frame.
  • the motion module 1105 is further configured to extract features of the image frame; acquire features of the node image corresponding to the start node; determine a spatial state difference between the feature of the image frame and the feature of the node image; The amount of state difference is exercised.
  • the current position of the node image corresponding to the node included in the map is similarly matched to locate the current position in the map, so that the positioning result is more accurate.
  • the motion module 1105 is further configured to sequentially acquire features of the node image corresponding to each node included in the target motion path; and sequentially determine the spatial state difference between the features of the acquired node image of the corresponding adjacent node. ; The motion is performed according to the spatial state difference amount determined in turn.
  • the amount of spatial state difference between the currently acquired image frame and the node image corresponding to the determined initial node is moved to the initial node in the map, thereby moving to the target according to the selected trending target moving path. Exercise ensures the accuracy of the exercise.
  • a computer readable storage medium having stored thereon computer readable instructions that, when executed by a processor, implement the steps of: acquiring an image frame; performing face detection on the image frame to obtain an image frame including a face image Determining a face image corresponding to a target node in the map; selecting a start node matching the image frame from the map; wherein, the feature of the image frame matches the feature of the node image corresponding to the start node; according to the start node And the target node selects a trend moving path in the path included in the map; and moves according to the selected moving path of the target.
  • the image frame may be automatically determined to be corresponding to the face image when the image frame includes the face image is detected.
  • the target node locates the position of the target in the map, and then selects the start matching the image frame from the map based on the matching relationship between the feature of the image frame and the feature of the node image corresponding to each node in the map.
  • the node locates the current position of the local machine in the map, and then selects the moving path of the target to move in the path included in the map according to the current node and the target node. In this way, the positioning in the map can be completed by the feature matching between the images, the environmental influence caused by the positioning of the sensing signal is avoided, and the accuracy of the motion control is improved.
  • the computer readable instructions further cause the processor to perform the steps of: inputting an image frame into a convolutional neural network model; acquiring a feature map of the plurality of network layer outputs included in the convolutional neural network model; Input memory neural network model; obtain the result of whether the image frame output by the memory neural network model includes a face image.
  • the computer readable instructions further cause the processor to perform the steps of: extracting facial feature data of the facial image; and querying the preset facial image that matches the facial image according to the facial feature data; The face image obtains the target identification result; the service type associated with the target identification result is determined.
  • the computer readable instructions also cause the processor to perform the step of providing a service trigger entry corresponding to the type of service.
  • the computer readable instructions further cause the processor to perform the steps of: selecting an image frame from the image frames acquired in time series; determining whether the feature of the selected image frame conforms to a feature of the preset node image; When the feature of the image frame conforms to the feature of the node image, the selected image frame is obtained as a node image; the acquired node image is determined to be a corresponding node in the map; and the determined node stores the acquired feature of the node image.
  • determining the node of the acquired node image in the map comprises: extracting features of the acquired node image; acquiring features of the node image corresponding to the existing node in the map; determining the acquired feature and the extracted feature Between the change matrix; according to the node and the change matrix, determine the node of the acquired node image in the map.
  • the computer readable instructions further cause the processor to perform the steps of: calculating a similarity between a feature of the node image corresponding to the existing node in the map, and a feature of the acquired node image; When the similarity between the feature of the node image corresponding to the node and the acquired feature of the node image exceeds the preset similarity threshold, the ring including the existing node is generated in the map according to the node corresponding to the acquired node image. path.
  • selecting a starting node that matches the image frame from the map comprises: extracting features of the image frame; acquiring features of the node image corresponding to the node included in the map; determining features of the image frame and features of the node image The similarity between the nodes; the node corresponding to the feature of the node image with the highest similarity is selected, and the starting node matching the image frame is obtained.
  • the computer readable instructions further cause the processor to perform the steps of: extracting features of the image frame; acquiring features of the node image corresponding to the starting node; determining a space between features of the image frame and features of the node image The amount of state difference; the motion is based on the amount of spatial state difference.
  • moving the path according to the selected trending target comprises: sequentially acquiring features of the node image corresponding to each node included in the moving path of the target; and sequentially determining the acquired features of the node image of the corresponding adjacent node.
  • the amount of spatial state difference; the motion is performed according to the spatial state difference amount determined in turn.
  • a computer device comprising a memory and a processor, the memory storing computer readable instructions, when the computer readable instructions are executed by the processor, causing the processor to perform the following steps: acquiring an image frame; performing face detection on the image frame When the image frame includes the face image, determining the corresponding target node of the face image in the map; selecting a starting node matching the image frame from the map; wherein, the feature of the image frame and the feature image of the node image corresponding to the starting node Matching; according to the starting node and the target node, selecting a moving path of the target in the path included in the map; and moving according to the selected moving path of the target.
  • the computer device After acquiring the image frame, the computer device automatically determines the target node corresponding to the face image in the map when the image frame includes the face image, and locates the position of the target in the map, and then The matching relationship between the feature of the image frame and the feature of the node image corresponding to each node in the map is based on, and the starting node matching the image frame can be selected from the map, and the current position of the local device in the map is located, and then according to The current node and the target node can select moving toward the target motion path in the path included in the map. In this way, the positioning in the map can be completed by the feature matching between the images, the environmental influence caused by the positioning of the sensing signal is avoided, and the accuracy of the motion control is improved.
  • the computer readable instructions further cause the processor to perform the steps of: inputting an image frame into a convolutional neural network model; acquiring a feature map of the plurality of network layer outputs included in the convolutional neural network model; Input memory neural network model; obtain the result of whether the image frame output by the memory neural network model includes a face image.
  • the computer readable instructions further cause the processor to perform the steps of: extracting facial feature data of the facial image; and querying the preset facial image that matches the facial image according to the facial feature data; The face image obtains the target identification result; the service type associated with the target identification result is determined.
  • the computer readable instructions also cause the processor to perform the step of providing a service trigger entry corresponding to the type of service.
  • the computer readable instructions further cause the processor to perform the steps of: selecting an image frame from the image frames acquired in time series; determining whether the feature of the selected image frame conforms to a feature of the preset node image; When the feature of the image frame conforms to the feature of the node image, the selected image frame is obtained as a node image; the acquired node image is determined to be a corresponding node in the map; and the determined node stores the acquired feature of the node image.
  • determining the node of the acquired node image in the map includes: extracting features of the acquired node image; acquiring features of the node image corresponding to the existing node in the map; determining the acquired feature and the extracted feature Between the change matrix; according to the node and the change matrix, determine the node of the acquired node image in the map.
  • the computer readable instructions further cause the processor to perform the steps of: calculating a similarity between a feature of the node image corresponding to the existing node in the map, and a feature of the acquired node image; When the similarity between the feature of the node image corresponding to the node and the acquired feature of the node image exceeds the preset similarity threshold, the ring including the existing node is generated in the map according to the node corresponding to the acquired node image. path.
  • selecting a starting node that matches the image frame from the map comprises: extracting features of the image frame; acquiring features of the node image corresponding to the node included in the map; determining features of the image frame and features of the node image The similarity between the nodes; the node corresponding to the feature of the node image with the highest similarity is selected, and the starting node matching the image frame is obtained.
  • the computer readable instructions further cause the processor to perform the steps of: extracting features of the image frame; acquiring features of the node image corresponding to the starting node; determining a space between features of the image frame and features of the node image The amount of state difference; the motion is based on the amount of spatial state difference.
  • moving the path according to the selected trending target comprises: sequentially acquiring features of the node image corresponding to each node included in the moving path of the target; and sequentially determining the acquired features of the node image of the corresponding adjacent node.
  • the amount of spatial state difference; the motion is performed according to the spatial state difference amount determined in turn.
  • a service robot includes a memory and a processor, wherein the memory stores computer readable instructions, and when the computer readable instructions are executed by the processor, the processor performs the following steps: acquiring an image frame; and performing face detection on the image frame When the image frame includes the face image, determining the corresponding target node of the face image in the map; selecting a starting node matching the image frame from the map; wherein, the feature of the image frame and the feature image of the node image corresponding to the starting node Matching; according to the starting node and the target node, selecting a moving path of the target in the path included in the map; and moving according to the selected moving path of the target.
  • the service robot After acquiring the image frame, the service robot automatically determines the target node corresponding to the face image in the map when the image frame includes the face image, and locates the position of the target in the map, and then The matching relationship between the feature of the image frame and the feature of the node image corresponding to each node in the map is based on, and the starting node matching the image frame can be selected from the map, and the current position of the local device in the map is located, and then according to The current node and the target node can select moving toward the target motion path in the path included in the map. In this way, the positioning in the map can be completed by the feature matching between the images, the environmental influence caused by the positioning of the sensing signals is avoided, and the accuracy of the motion control is improved.
  • the computer readable instructions further cause the processor to perform the steps of: inputting an image frame into a convolutional neural network model; acquiring a feature map of the plurality of network layer outputs included in the convolutional neural network model; Input memory neural network model; obtain the result of whether the image frame output by the memory neural network model includes a face image.
  • the computer readable instructions further cause the processor to perform the steps of: extracting facial feature data of the facial image; and querying the preset facial image that matches the facial image according to the facial feature data; The face image obtains the target identification result; the service type associated with the target identification result is determined.
  • the computer readable instructions also cause the processor to perform the step of providing a service trigger entry corresponding to the type of service.
  • the computer readable instructions further cause the processor to perform the steps of: selecting an image frame from the image frames acquired in time series; determining whether the feature of the selected image frame conforms to a feature of the preset node image; When the feature of the image frame conforms to the feature of the node image, the selected image frame is obtained as a node image; the acquired node image is determined to be a corresponding node in the map; and the determined node stores the acquired feature of the node image.
  • determining the node of the acquired node image in the map includes: extracting features of the acquired node image; acquiring features of the node image corresponding to the existing node in the map; determining the acquired feature and the extracted feature Between the change matrix; according to the node and the change matrix, determine the node of the acquired node image in the map.
  • the computer readable instructions further cause the processor to perform the steps of: calculating a similarity between a feature of the node image corresponding to the existing node in the map, and a feature of the acquired node image; When the similarity between the feature of the node image corresponding to the node and the acquired feature of the node image exceeds the preset similarity threshold, the ring including the existing node is generated in the map according to the node corresponding to the acquired node image. path.
  • selecting a starting node that matches the image frame from the map comprises: extracting features of the image frame; acquiring features of the node image corresponding to the node included in the map; determining features of the image frame and features of the node image The similarity between the nodes; the node corresponding to the feature of the node image with the highest similarity is selected, and the starting node matching the image frame is obtained.
  • the computer readable instructions further cause the processor to perform the steps of: extracting features of the image frame; acquiring features of the node image corresponding to the starting node; determining a space between features of the image frame and features of the node image The amount of state difference; the motion is based on the amount of spatial state difference.
  • moving the path according to the selected trending target comprises: sequentially acquiring features of the node image corresponding to each node included in the moving path of the target; and sequentially determining the acquired features of the node image of the corresponding adjacent node.
  • the amount of spatial state difference; the motion is performed according to the spatial state difference amount determined in turn.
  • Non-volatile memory can include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in a variety of formats, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronization chain.
  • SRAM static RAM
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • DDRSDRAM double data rate SDRAM
  • ESDRAM enhanced SDRAM
  • Synchlink DRAM SLDRAM
  • Memory Bus Radbus
  • RDRAM Direct RAM
  • DRAM Direct Memory Bus Dynamic RAM
  • RDRAM Memory Bus Dynamic RAM

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • General Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Human Computer Interaction (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Automation & Control Theory (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Electromagnetism (AREA)
  • Image Analysis (AREA)

Abstract

一种运动控制方法，包括：获取图像帧；当对所述图像帧进行人脸检测得到所述图像帧包括人脸图像时，确定所述人脸图像在地图中相应的目标节点；从所述地图中挑选与所述图像帧匹配的起始节点；其中，所述图像帧的特征与所述起始节点对应的节点图像的特征相匹配；根据所述起始节点和所述目标节点，在所述地图包括的路径中选取趋向目标运动路径；按照选取的所述趋向目标运动路径运动。

Description

运动控制方法、存储介质、计算机设备和服务机器人
本申请要求于2017年05月22日提交中国专利局，申请号为201710365516.X，申请名称为“运动控制方法、装置、计算机设备和服务机器人”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域，特别是涉及一种运动控制方法、存储介质、计算机设备和服务机器人。
背景技术
随着计算机技术的发展和人们生活水平的提高，人们越来越依赖于可移动的计算机设备来帮助人们完成各种任务。在传统的通过可移动的计算机设备在执行任务时，可移动的计算机设备的运动控制是基于传感器的定位方式来实现的。
然而，基于传统的这种通过传感器的定位方式来控制计算机设备的运动时，在定位过程中传感信号容易受到周围环境的影响，会严重影响定位的准确度，从而导致在控制计算机设备的运动时准确率降低。
发明内容
根据本申请提供的各种实施例，提供一种运动控制方法、存储介质、计算机设备和服务机器人。
一种运动控制方法，包括：
计算机设备获取图像帧；
所述计算机设备当对所述图像帧进行人脸检测得到所述图像帧包括人脸图像时，确定所述人脸图像在地图中相应的目标节点；
所述计算机设备从所述地图中挑选与所述图像帧匹配的起始节点；其中，所述图像帧的特征与所述起始节点对应的节点图像的特征相匹配；
所述计算机设备根据所述起始节点和所述目标节点，在所述地图包括的路径中选取趋向目标运动路径；
所述计算机设备按照选取的所述趋向目标运动路径运动。
一个或多个存储有计算机可读指令的非易失性存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行以下步骤：
获取图像帧；
当对所述图像帧进行人脸检测得到所述图像帧包括人脸图像时，确定所述人脸图像在地图中相应的目标节点；
从所述地图中挑选与所述图像帧匹配的起始节点；其中，所述图像帧的特征与所述起始节点对应的节点图像的特征相匹配；
根据所述起始节点和所述目标节点，在所述地图包括的路径中选取趋向目标运动路径；及
按照选取的所述趋向目标运动路径运动。一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行以下步骤：
获取图像帧；
当对所述图像帧进行人脸检测得到所述图像帧包括人脸图像时，确定所述人脸图像在地图中相应的目标节点；
从所述地图中挑选与所述图像帧匹配的起始节点；其中，所述图像帧的特征与所述起始节点对应的节点图像的特征相匹配；
根据所述起始节点和所述目标节点，在所述地图包括的路径中选取趋向目标运动路径；及
按照选取的所述趋向目标运动路径运动。
一种服务机器人，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行以 下步骤：
获取图像帧；
当对所述图像帧进行人脸检测得到所述图像帧包括人脸图像时，确定所述人脸图像在地图中相应的目标节点；
从所述地图中挑选与所述图像帧匹配的起始节点；其中，所述图像帧的特征与所述起始节点对应的节点图像的特征相匹配；
根据所述起始节点和所述目标节点，在所述地图包括的路径中选取趋向目标运动路径；及
按照选取的所述趋向目标运动路径运动。
本申请的一个或多个实施例的细节在下面的附图和描述中提出。本申请的其它特征、目的和优点将从说明书、附图以及权利要求书变得明显。
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，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的附图。
图1为一个实施例中运动控制方法的应用环境图；
图2为一个实施例中用于实现运动控制方法的计算机设备的内部结构图；
图3为一个实施例中运动控制方法的流程示意图；
图4为一个实施例中人脸检测的步骤的流程示意图；
图5为一个实施例中对人脸图像进行人脸识别的示意图；
图6为一个实施例中构建地图的步骤的流程示意图；
图7为一个实施例中地图创建过程的流程示意图；
图8为一个实施例中创建完成的地图的示意图；
图9为一个实施例中在地图中选取趋向目标运动路径的示意图；
图10为另一个实施例中运动控制方法的流程示意图；
图11为一个实施例中计算机设备的结构框图；
图12为另一个实施例中计算机设备的结构框图；
图13为又一个实施例中计算机设备的结构框图；及
图14为再一个实施例中计算机设备的结构框图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
图1为一个实施例中运动控制方法的应用环境图。如图1所示，该运动控制方法应用于运动控制系统。该运动控制系统应用于室内场景。运动控制系统包括计算机设备110和目标120。计算机设备110可通过执行运动控制方法，向目标120运动。本领域技术人员可以理解，图1中示出的应用环境，仅仅是与本申请方案相关的部分场景，并不构成对本申请方案应用环境的限定，该运动控制系统还可应用于室外开阔场景中等。
图2为一个实施例中计算机设备的内部结构示意图。如图2所示，该计算机设备包括通过系统总线连接的处理器、存储器、摄像头、声音采集装置、扬声器、显示屏、输入装置和运动装置。其中，存储器包括非易失性存储介质和内存储器。该计算机设备的非易失性存储介质存储有操作系统，还可存储有计算机可读指令，该计算机可读指令被处理器执行时，可使得处理器实现一种运动控制方法。该处理器用于提供计算和控制能力，支撑整个计算机设备的运行。该内存储器中也可储存有计算机可读指令，该计算机可读指令被所述处理器执行时，可使得所述处理器执行一种运动控制方法。计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏等，输入装置可以是显示屏上覆盖的触摸层，也可以是终端外壳上设置的按键、轨迹球或触控板，也可以是外接的键盘、触控板或鼠标等。该计算机设备是可移动的电子设备， 具体可以是服务机器人等。本领域技术人员可以理解，图2中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的终端的限定，具体的终端可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
如图3所示，在一个实施例中，提供了一种运动控制方法。本实施例主要以该方法应用于上述图2中的计算机设备来举例说明。参照图3，该运动控制方法具体包括如下步骤：
S302，获取图像帧。
在一个实施例中，计算机设备可通过摄像头，在摄像头当前的视野下采集图像帧，获取采集得到的图像帧。其中，摄像头的视野可因计算机设备的姿态和位置的变化而变化。
在一个实施例中，计算机设备具体可按照固定或动态的帧率采集图像帧，获取采集得到的图像帧。其中，固定或动态的帧率能够使图像帧按照该固定或动态的帧率播放时形成连续的动态画面，以使计算机设备可追踪连续的动态画面中的特定目标。
在一个实施例中，计算机设备可调用摄像头开启摄像扫描模式，实时扫描当前的视野下的特定目标，并按照一定的帧率实时地生成图像帧，获取生成的图像帧。
其中，计算机设备是可移动的电子设备，具体可以是机器人等。摄像头可以是计算机设备内置的摄像头，或者外置的与计算机设备关联的摄像头。摄像头可以是单目摄像头、双目摄像头或者RGB-D(Red-Green-Blue-Deep)摄像头等。
S304，当对图像帧进行人脸检测得到图像帧包括人脸图像时，确定人脸图像在地图中相应的目标节点。
其中，地图是计算机设备根据从自然空间中采集的图像帧构建的特征分布图。计算机设备可基于SLAM(Simultaneous Localization And Mapping定位与地图构建)对自然空间构建相应的地图。计算机设备基于SLAM构建的 地图具体可以是三维点图。节点是计算机设备将从自然空间中采集图像帧的位置投影至地图空间中的位置。目标节点是目标在自然空间中的位置投影至地图的节点。比如，目标在自然空间的坐标为A(x1，y1，z1)，将A投影至地图空间后坐标为B(x2，y2，z2)，那么B即为目标在地图中的节点。
在一个实施例中，计算机设备可在获取到图像帧后，提取该图像帧中包括的图像数据，并检测该图像数据是否包含人脸特征数据。若计算机设备检测到该图像数据中包含人脸特征数据，则判定该图像帧中包括人脸图像。计算机设备也可在获取到图像帧后，将该图像帧发送至服务器，由服务器完成对图像帧的人脸检测过程，再向计算机设备返回图像帧中是否包括人脸图像的检测结果。其中，检测结果可包括图像帧中存在人脸图像的概率和人脸图像的坐标区域。
在一个实施例中，地图中可包括若干节点，各节点均存在一一对应的节点图像。地图还可包括从节点图像中提取出的特征点。包括特征点和节点的地图是对自然空间中场景的三维重建。具体地，自然空间中的三维场景中的三维点通过投影矩阵的投影变换，得到计算机设备摄像头摄像平面的二维图像帧中的像素点，二维图像帧中的像素点再经过投影矩阵的投影反变换，得到地图中的三维重建场景中的三维特征点。
计算机设备可在检测到该图像帧中包括人脸图像时，计算该人脸图像在地图中的位置。具体地，计算机设备可确定人脸图像在该图像帧中的坐标位置，根据与计算机设备的摄像头适配的投影矩阵，计算该人脸图像在地图中的位置，在地图中包括的节点中查找与计算得到的位置相应的节点，得到目标节点。
在一个实施例中，计算机设备可在检测到该图像帧中包括人脸图像时，提取该图像帧中背景图像的背景特征点，将提取的背景特征点与地图中包括的特征点进行匹配，获取地图中与提取的背景特征点匹配的特征点的位置，从而在地图中选取与该位置的距离最近的节点，得到目标节点。
在一个实施例中，计算机设备获取的图像帧可以是两帧或者两帧以上的 图像帧。计算机设备在检测得到获取的图像中存在人脸图像时，可计算任意两帧图像帧之间的相似矩阵，再从用于计算相似矩阵的图像帧中包括的人脸图像上选取匹配的人脸特征点，并确定该人脸特征点在图像帧上的位置。计算机设备可再根据计算得到的任意两帧图像帧之间的相似矩阵以及选取的人脸特征点在这两帧图像上的位置，按照三角测距算法确定该人脸特征点在自然空间中的位置。计算机设备可再根据该人脸特征点在自然空间中的位置确定该人脸特征点在地图中的位置，从而在地图中选取与该位置的距离最近的节点，得到目标节点。
S306，从地图中挑选与图像帧匹配的起始节点；其中，图像帧的特征与起始节点对应的节点图像的特征相匹配。
其中，节点图像是计算机设备在与地图中的节点存在投影关系的自然空间中的位置处采集的图像。图像的特征可以是颜色特征、纹理特征和形状特征中的一种或几种的组合。计算机设备可在构建地图时，对地图中节点对应的节点图像提取特征，将提取的节点图像的特征相对于相应的节点存储在数据库或者缓存中。
在一个实施例中，计算机设备可遍历地图中各节点对应的节点图像的特征，判断遍历至的节点图像的特征与图像帧的特征是否匹配。计算机设备可在判定遍历至的节点图像的特征与图像帧的特征匹配时，获取遍历至的节点图像的特征所对应的节点为起始节点。
在一个实施例中，计算机设备在判断遍历至的节点图像的特征与图像帧的特征是否匹配时，具体可先计算遍历至的节点图像的特征与图像帧的特征之间的相似度，进而判断该相似度是否大于等于预设相似度；若是，则匹配；若否，则不匹配。其中，相似度可采用余弦相似度或者图像间各自感知哈希值的汉明距离。
在一个实施例中，计算机设备具体可根据节点图像中各像素点的像素值，选取极值点作为特征点。其中，计算机可基于FAST(Features from accelerated segment test快速特征点检测)或者Harris角点检测算法等算法选取极值点， 得到节点图像的特征点，再将得到的特征点通过二进制编码表示。计算机设备可再用一维图像特征向量表示节点图像包括的特征点，得到与地图的节点一一对应的一维图像特征向量。
计算机设备可按照表征节点图像的特征的方式，生成表征获取的图像帧的特征的一维图像特征向量。计算机设备可再计算生成的一维图像特征向量与地图的各节点对应的一维图像特征向量之间的向量相似度，进而判断该向量相似度是否大于等于预设向量相似度；若是，则匹配；若否，则不匹配。
S308，根据起始节点和目标节点，在地图包括的路径中选取趋向目标运动路径。
具体地，地图中可包括通过地图中的节点形成的路径。计算机设备可以起始节点作为起点，目标节点为终点，在地图中通过节点形成的路径中选取路径得到趋向目标运动路径。
在一个实施例中，地图以起始节点作为起点，目标节点为终点的路径可以是一条或者多条。当以起始节点作为起点，目标节点为终点的路径唯一时，计算机设备可直接获取该路径为趋向目标运动路径。当以起始节点作为起点，目标节点为终点的路径不唯一时，计算机设备可随机选取一条路径作为趋向目标运动路径，也可获取包括的节点数最少的路径作为趋向目标运动路径。
S310，按照选取的趋向目标运动路径运动。
具体地，计算机设备在选取趋向目标运动路径运动后，获取该路径所包括的各节点对应的节点图像的特征，按照各节点对应的节点图像的特征之间的变化关系确定计算机设备当前运动的方向和距离，按照确定的方向和距离向目标运动。
上述运动控制方法，在获取到图像帧后，就可以自动地在检测到该图像帧包括人脸图像时，在地图中确定该人脸图像相应的目标节点，定位目标在地图中的位置，然后以该图像帧的特征与地图中各节点对应的节点图像的特征的匹配关系为依据，即可从地图中挑选与该图像帧匹配的起始节点，定位本机当前在地图中的位置，再根据当前节点和目标节点便可在地图包括的路 径中选取趋向目标运动路径来运动。这样通过图像之间的特征匹配即可完成在地图中的定位，避免了通过传感信号定位引起的环境影响，提高了运动控制的准确性。
在一个实施例中，该运动控制方法还包括人脸检测的步骤。可以理解，人脸检测的步骤可以在S302之后执行。该人脸检测的步骤具体包括：
S402，将图像帧输入卷积神经网络模型。
其中，卷积神经网络模型是由多层互相连接而形成的复杂网络模型。神经网络模型可包括多层特征转换层，每层特征转换层都有对应的非线性变化算子，每层的非线性变化算子可以是多个，每层特征转换层中一个非线性变化算子对输入的图像进行非线性变化，得到特征图(Feature Map)作为运算结果。
具体地，卷积神经网络模型是以包括人脸图像的图像作为训练数据，进行学习训练得到的用于提取人脸特征的模型。计算机设备在获取到图像帧后，将图像帧输入卷积神经网络模型，利用卷积神经网络模型对图像帧进行人脸特征提取。其中，人脸特征可以是用于反映出人的性别、人脸的轮廓、发型、眼镜、鼻子、嘴以及各个脸部器官之间的距离等其中的一种或多种特征。
在一个实施例中，卷积神经网络模型是以图像作为训练数据，进行学习训练得到的用于提取图像特征的模型。计算机设备在获取到图像帧后，将图像帧输入卷积神经网络模型，利用卷积神经网络模型对图像帧进行图像特征提取。
S404，获取卷积神经网络模型包括的多个网络层输出的特征图。
具体地，计算机设备可获取卷积神经网络模型包括的多个网络层输出的特征图。特征图是由非线性变化算子对输入的图像进行处理得到的响应值构成的。不同的网络层提取的特征不同。计算机设备可利用提取人脸特征的卷积神经网络输出的特征图确定输入的图像相应的人脸特征数据。计算机设备可利用提取图像特征的卷积神经网络输出的特征图确定输入的图像相应的图像特征数据，进而判断该图像特征数据中是否包括人脸特征数据。
举例说明，计算机设备可采用52层深度残差网络模型进行图像处理，提取该深度残差网络模型中包括的4层全连接层输出的特征图，作为后续输入。
S406，将各特征图依次输入记忆神经网络模型。
其中，记忆神经网络模型是可对序列输入进行综合处理的神经网络模型。记忆神经网络模型是递归神经网络模型。记忆神经网络模型具体可以是LSTM(Long Short-Term Memory长短时记忆神经网络)。具体地，计算机设备可将获取的各特征图依次输入记忆神经网络模型，进行人脸特征检测。
S408，获取记忆神经网络模型输出的图像帧是否包括人脸图像的结果。
具体地，计算机设备可获取记忆神经网络模型根据输入的各特征图综合处理得到的人脸检测结果。人脸检测结果包括存在人脸图像的概率和人脸图像在图像帧中的坐标区域。
在一个实施例中，计算机设备还可在提取得到人脸检测结果后，根据人脸检测结果中包括的人脸图像在图像帧中的坐标区域，过滤掉重叠区域超过预设重叠阈值的人脸检测结果，根据过滤后保留的人脸检测结果得到人脸图像在图像帧中的坐标区域。
在一个实施例中，记忆神经网络模型可使用一个矩形窗口，按照预设方向和预设步长在输入的特征图中移动，从而进行窗口扫描，在扫描时提取扫描至的窗口图像中人脸特征数据，根据提取的人脸特征图像，得到扫描至的窗口图像中存在人脸图像的概率。将计算得到的概率排序靠前的窗口图像在图像帧中的坐标区域进行存储，并继续对后续输入的特征图进行处理。
图5示出了一个实施例中对人脸图像进行人脸识别的示意图。参考图5，计算机设备采用的记忆神经网络模型对输入的特征图按照矩形窗口扫描分析，得到与矩形窗口A对应的存在人脸图像的概率P A，矩形窗口B对应的存在人脸图像的概率P B，矩形窗口C对应的存在人脸图像的概率P C。此时，P C>P A>P B，记忆神经网络模型可将P C对应的矩形窗口C进行记录，继续对后续输入的特征图按照矩形窗口扫描分析，并综合多次分析得到矩形窗口以及相应的存在人脸图像的概率，输出计算机设备获取的图像帧中存在人脸图像 的概率以及该人脸图像在图像帧中的坐标区域。
在本实施例中，通过卷积神经网络模型的包括的多个网络层充分提取图像特征，再将多层网络层提取的特征输入记忆神经网络模型综合处理，使得人脸检测更准确。
在一个实施例中，该运动控制方法还包括人脸识别的步骤，人脸识别的步骤具体包括：提取人脸图像的人脸特征数据；根据人脸特征数据查询与人脸图像相匹配的预设人脸图像；根据预设人脸图像得到目标身份识别结果；确定与目标身份识别结果相关联的服务类型。其中，上述人脸识别的步骤具体可在S304之后执行。该运动控制方法还包括：提供与服务类型相应的服务触发入口。该步骤具体可在S310之后执行。
其中，目标身份识别结果是用于反映目标身份的数据。目标身份可以是目标的名字、社会地位或者职位信息等。
在一个实施例中，计算机设备上设置有预设人脸图像库，预设人脸图像库中包括若干预设人脸图像。计算机设备可在检测到图像帧中包括人脸图像时，将图像帧中的人脸图像与预设人脸图像库中包括的预设人脸图像比较，检测图像帧中的人脸图像和预设人脸图像之间是否匹配。计算机设备可在图像帧中的人脸图像和预设人脸图像之间匹配时，判定该图像帧包括的人脸图像与预设人脸图像为相同的人物图像，获取该预设人脸图像对应的目标身份信息作为目标身份识别结果。
其中，预设人脸图像可以是用于反映对应目标的真实人脸图像。可从目标所上传的个人资料、历史发表的图片信息中，由对应目标自定义选取的图像，或由系统自动地分析选取的一张图片，作为相应的预设人脸图像。
在一个实施例中，计算机设备在检测图像帧中的人脸图像和预设人脸图像之间是否匹配，具体可计算图像帧中的人脸图像和预设人脸图像之间的相似度。计算机设备可先提取图像帧中的人脸图像和预设人脸图像各自的特征，从而计算两特征之间的差异，特征之间的差异越大则相似度越低，特征之间的差异越小则相似度越高。其中，计算机设备计算图像帧中的人脸图像和预 设人脸图像之间的相似度时，可以采用适于图像处理器的加速算法，提高运算速率。
在一个实施例中，计算机设备可在判定该图像帧中包括人脸图像后从该图像数据中提取人脸特征数据，再将提取的人脸特征数据与预设人脸图像库中各预设人脸图像相对应的人脸特征数据比较，得到目标身份识别结果。
在一个实施例中，计算机设备对图像帧进行检测得到的该图像帧包括的人脸图像可以是一个或者多个。计算机设备可确定图像帧中包括的人脸图像占图像帧的占比，提取占比超过预设比例的人脸图像的人脸特征数据；和/或，确定图像帧中包括的人脸图像的清晰度，提取清晰度超过清晰度阈值的人脸图像的人脸特征数据。计算机设备再对提取了人脸特征数据的人脸图像进行识别。
进一步地，计算机设备在识别得到目标身份识别结果后，可查找与目标身份识别结果相关联的服务类型。其中，服务类型是向目标提供的服务所属的类型。服务类型比如餐厅点餐服务或者酒店接待服务等。服务类型可以是统一设置的类型，也可以是与目标身份相关的类型，还可以是与目标属性相关的类型。
在一个实施例中，计算机设备可事先设置服务类型，并将服务类型与目标标识关联，再将设置的服务类型存储在数据库或者文件中，在需要时从数据库或者文件中读取。计算机设备在识别得到目标身份识别结果后，可拉取该目标身份识别结果对应的目标标识所关联的服务类型。
更进一步地，计算机设备在确定与目标身份识别结果相关联的服务类型后，可在运动至目标后，向目标提供与确定的服务类型相应的服务触发入口。具体地，计算机设备可通过显示屏提供服务触发入口，也可通过扬声器和声音采集器与目标提供语音服务入口。
在一个实施例中，计算机设备在运动至目标节点处后，可采集图像帧确定当前所在位置，并为目标提供服务触发入口，通过显示屏或声音采集器接收输入的服务参数，计算机设备从而确定当前服务的对象、当前服务的位置 以及当前服务的内容。
上述实施例中，在对获取的图像检测到存在人脸时，对存在的人脸进行识别，在识别得到目标的身份并运动至目标后即可向该目标提供与该目标相关联的服务入口，极大地提高了服务提供的效率。
上述实施例中，由计算机设备处理的人脸识别步骤和确定与目标身份识别结果相关联的服务类型均可由服务器处理。计算机设备可将获取的图像帧发送至服务器，服务器在对图像帧完成人脸检测、人脸识别以及确定与目标身份识别结果相关联的服务类型后，将目标身份识别结果与相关联的服务类型发送至计算机设备。
在一个实施例中，该运动控制方法还包括构建地图的步骤。可以理解构建地图的步骤具体可在S402之前执行。该构建地图的步骤具体包括：
S602，从按时序采集的图像帧中选取图像帧。
其中，选取的图像帧，可以是采集的图像帧中的关键帧。
在一个实施例中，计算机设备可接收用户选择指令，根据该用户选择指令，从采集的图像帧中选取图像帧。
在一个实施例中，计算机设备可按照预设间隔帧数从采集的图像帧中选取图像帧。比如，每隔20帧图像帧后选取图像帧。
S604，判断选取的图像帧的特征是否符合预设的节点图像的特征。
具体地，预设的节点图像的特征是预设的用于选择节点图像的特征。符合预设的节点图像的特征可以是图像中包括的特征点与已有节点图像包括的特征点中相匹配的特征点的数量超过预设数量，也可以是包括的特征点与已有节点图像包括的特征点中相匹配特征点占已有节点图像包括的特征点的比例低于预设比例。
举例说明，假设最近添加的节点图像包括的特征点数量为100，当前选取的图像帧包括的特征点数量为120。预设数量为50，预设比例为90％。其中，若当前选取的图像帧包括的特征点与最近添加的节点图像包括的特征点中相匹配的特征点的数量为70。那么，当前图像帧中包括的特征点与已有节 点图像包括的特征点匹配的数量超过预设数量，可判定当前选取的图像帧的特征符合预设的节点图像的特征。
S606，当选取的图像帧的特征符合节点图像的特征时，获取选取的图像帧为节点图像。
在一个实施例中，计算机设备在获取构建地图的指令后，可按照固定或动态的帧率采集图像帧，选取采集的图像帧包括的特征点的数量大于预设数量阈值的图像帧为初始的节点图像，确定该节点图像在地图中相应的节点，以及该节点图像包括的特征点在地图中相应的位置，构建局部地图。计算机设备再从按时序采集的图像帧中选取图像帧，将选取符合预设的节点图像的特征的图像帧作为后续的节点图像，直至得到全局地图。
具体地，计算机设备可以初始的节点图像为参考节点图像，追踪参考节点图像中的特征点。当选取的图像帧包括的特征点与参考节点图像包括的特征点的匹配数量低于第一预设数量且高于第二预设数量时，将选取的图像帧作为节点图像。当选取的图像帧包括的特征点与参考节点图像包括的特征点的匹配数量低于第二预设数量时，将最近获取的节点图像为参考节点图像，继续进行图像追踪，以选取节点图像。
S608，确定获取的节点图像在地图中相应的节点。
具体地，计算机设备可确定在自然空间中采集该获取的节点图像投影于地图空间中的节点。计算机设备可提取在获取的节点图像时序靠前的节点图像的特征，计算时序靠前的节点图像的特征与获取的节点图像的变化矩阵，根据该变化矩阵得到采集时序靠前的节点图像时的位置到采集获取的节点图像时的位置的变化量，再根据该变化量确定获取的节点图像在地图中相应的节点。
在一个实施例中，步骤S608包括：提取获取的节点图像的特征；获取地图中已有的节点对应的节点图像的特征；确定获取的特征与提取的特征之间的变化矩阵；根据节点与变化矩阵，确定获取的节点图像在地图中相应的节点。
其中，变化矩阵是二维图像的特征到二维图像的特征之间的相似变化关系。具体地，计算机设备可提取获取的节点图像的特征，地图中已有的节点对应的节点图像的特征进行匹配，获取匹配成功的特征分别在获取的节点图像和已有的节点图像中的位置。获取的节点图像为在后采集的图像帧，已有的节点图像为在后采集的图像帧。计算机设备即可根据得到的匹配的特征在先后采集的两帧图像帧上的位置确定先后采集的两帧图像帧之间的变化矩阵，从而得到计算机设备采集这两帧图像帧时的位置变化和姿态变化，再根据在前采集的图像的位置和姿态，即可得到在后采集的图像的位置和姿态。
在一个实施例中，地图中已有的节点对应的节点图像可以是一帧或者多帧。计算机设备也可将获取的节点图像的特征与多个已有的节点对应的节点图像的特征比较，得到在采集后的图像帧与多个在先采集的图像帧的变化矩阵，再根据多个变化矩阵综合得到在后采集的图像的位置和姿态。比如，对计算得到的多个位置变化和姿态变化加权求平均等。
在本实施例中，通过节点图像的特征之间的变化矩阵，得到当前获取的节点图像与在前已有的节点图像的转化关系，从而实现由在前的图像帧在地图中的位置推测当前图像帧的在地图中的位置，实现实时定位。
S610，对应于确定的节点存储获取的节点图像的特征。
具体地，计算机设备可提取节点图像的特征，将节点图像的特征对应于节点图像相应的节点存储，可在需要进行图像特征比较时，直接根据节点查找对应的节点图像的特征，以节省存储空间提高查找效率。
在本实施例中，通过自身采集图像帧，再对采集的图像帧进行处理即可自动进行地图构建，避免了需要大量具备专业绘图能力的工作人员人工对环境进行测绘，对工作人员能力要求高且劳动量大的问题，提高地图构建的效率。
在一个实施例中，该运动控制方法还包括：计算地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度；当地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度超过 预设相似度阈值时，则根据获取的节点图像相应的节点，在地图中生成包括已有的节点的环形路径。可以理解，这些步骤具体可在S608之后执行。
具体地，计算机设备获取节点图像时，可将新增的该节点图像的特征与地图中已有的节点对应的节点图像的特征进行比较，计算新增的节点图像的特征与地图中已有的节点对应的节点图像的特征之间的相似度。当地图中已有的节点对应的节点图像的特征，与新增的节点图像的特征之间的相似度超过预设相似度阈值时，计算机设备可判定新增的节点图像在自然空间中的采集位置与已有的节点对应的节点图像在自然空间中的采集位置一致。
计算机设备可通过获取的节点图像相应的节点，在地图中生成自该已有的节点起，经过该已有的节点后添加的节点，至自该已有的节点的环形路径。计算机设备可再自该已有的节点起依次顺序获取环形路径包括的各节点所对应的节点图像的特征；依次确定获取的对应相邻节点的节点图像的特征之间的变化矩阵；根据依次确定的变化矩阵逆序调整环形路径包括的各节点所对应的节点图像的特征。
举例说明，计算机设备从第一帧节点图像起，依次添加节点图像构建局部地图。在检测到当前第四帧节点图像的特征与第一帧节点图像的特征之间的相似度超过预设相似度阈值时，判定第四帧节点图像在自然空间中的采集位置与第一帧节点图像在自然空间中的采集位置一致，生成第一帧节点图像-第二帧节点图像-第三帧节点图像-第一帧节点图像的环形路径。
其中，第一帧节点图像的特征与第二帧节点图像的特征之间的变化矩阵为H1，第二帧节点图像的特征与第三帧节点图像的特征之间的变化矩阵为H2，第三帧节点图像的特征与第四帧节点图像的特征之间的变化矩阵为H4。计算机设备可将第一帧节点图像的特征按照H4变化，根据得到的图像的特征优化第三帧节点图像，再将优化后的第三帧节点图像按照H3变化，根据得到的图像的特征优化第二帧节点图像。
在本实施例中，以新增的节点图像的特征与已有的节点图像的特征的相似度作为依据进行闭环检测，在检测到有闭环时，在地图中生成环形路径， 以进行后续的闭环优化，提高构建地图的准确性。
图7示出了一个实施例中地图创建过程的流程示意图。参考图7，该地图创建过程包括追踪、建图和闭环检测三个部分。计算机设备在获取构建地图的指令后，可按照固定或动态的帧率采集图像帧。在采集到图像帧后，提取该图像帧的特征点，将提取的特征点与地图中新增的节点对应的节点图像的特征点匹配。当提取的特征点与地图中新增的节点对应的节点图像的特征点匹配失败时，计算机设备可重新获取采集的图像帧进行重定位。
当提取的特征点与地图中新增的节点对应的节点图像的特征点匹配成功时，根据地图中新增的节点预估采集的该图像帧对应与地图中的节点。计算机设备可再追踪地图中与采集的该图像相匹配的特征点，根据相匹配的特征优化该图像帧对应与地图中的节点。在对采集的该图像优化完成后，判断该图像帧的特征点是否符合预设的节点图像的特征点，若否，计算机设备可重新获取采集的图像帧进行特征点匹配。
若该图像帧的特征点符合预设的节点图像的特征点，计算机设备可获取该图像帧为新增的节点图像。计算机设备可提取该新增的节点图像的特征点，按照预设的统一的格式表示提取的特征点，再按照三角测距算法确定新增的节点图像的特征点在地图中的位置，从而更新局部地图，再进行局部集束调整，去除相似度高于预设相似度阈值的节点图像对应的冗余的节点。
计算机设备在获取该图像帧为新增的节点图像后，可异步进行闭环检测。将新增的节点图像的特征与已有的节点对应的节点图像的特征进行对比，当新增的节点图像的特征与已有的节点对应的节点图像的特征之间的相似度高于预设相似度阈值，计算机设备可判定新增的节点图像在自然空间中的采集位置与已有的节点对应的节点图像在自然空间中的采集位置一致，即存在闭环。计算机设备可再根据新增的节点图像相应的节点，在地图中生成包括位置一致的节点的环形路径，并进行闭环优化和闭环融合。最终得到包括特征点、节点和路径的全局地图
图8示出了一个实施例中创建完成的地图的示意图。参考图8，该地图 是基于稀疏特征建立的特征分布示意图。该示意图包括特征点801、节点802以及节点间形成的路径803。其中，特征点801是自然空间中物体的特征点在自然空间中的位置在地图空间中的投影位置。节点802是计算机设备在自然空间中采集图像帧时的自然空间位置在地图空间的投影位置。节点间形成的路径803是计算机设备在自然空间中运动的路径在地图空间中的投影。
在一个实施例中，步骤S306包括：提取图像帧的特征；获取地图包括的节点所对应的节点图像的特征；确定图像帧的特征与节点图像的特征之间的相似度；选取对应相似度最高的节点图像的特征所对应的节点，得到与图像帧相匹配的起始节点。
具体地，计算机设备在将计算地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征比较时，可计算两图像特征之间的差异，特征之间的差异越大则相似度越低，特征之间的差异越小则相似度越高。相似度可采用余弦相似度或者图像间各自感知哈希值的汉明距离。计算机设备在计算得到地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度后，选取对应相似度最高的节点图像的特征所对应的节点，得到与图像帧相匹配的起始节点。
在本实施例中，通过当前图像帧与地图包括的节点所对应的节点图像的特征相似匹配来定位当前在地图中的位置，使得自身定位结果更加准确。
在一个实施例中，该运动控制方法还包括：提取图像帧的特征；获取起始节点所对应的节点图像的特征；确定图像帧的特征和节点图像的特征之间的空间状态差异量；根据空间状态差异量进行运动。可以理解，这些步骤可在S310之前执行。
其中，空间状态差异量是计算机设备在采集不同的图像帧时空间状态的变化量。空间状态差异量包括空间位置差异量和空间角度差异量。空间位置差异量是计算机设备在物理位置上的移动。比如，计算机设备在采集第一帧图像帧时至采集第二帧图像帧时水平向前平移0.5m。空间角度差异量是计算机设备在物理方位上的旋转，比如，计算机设备在采集第一帧图像帧时至采 集第二帧图像帧时逆时针旋转15度。
具体地，计算机设备可计算图像帧的特征与起始节点所对应的节点图像的特征之间的变化矩阵，根据计算得到的变化矩阵恢复计算机设备的运动位置，从变化矩阵中分解得到旋转矩阵和位移矩阵，根据旋转矩阵得到图像帧的特征和节点图像的特征之间的空间角度差异量，根据位移矩阵得到图像帧的特征和节点图像的特征之间的空间位置差异量。计算机设备可再根据空间角度差异量确定当前运动的方向，根据空间位置差异量确定当前运动的距离，从而按照确定的方向移动确定的距离。
在本实施例中，通过当前获取到的图像帧与确定的初始节点对应的节点图像之间的空间状态差异量，以运动至地图中的初始节点处，从而按照选取的趋向目标运动路径向目标运动，保证了运动的准确性。
在一个实施例中，步骤S310包括：依次获取趋向目标运动路径包括的各节点所对应的节点图像的特征；依次确定获取的对应相邻节点的节点图像的特征之间的空间状态差异量；根据依次确定的空间状态差异量进行运动。
具体地，计算机设备可获取趋向目标运动路径包括的与初始节点相邻的第二节点所对应的节点图像的特征，计算初始节点对应的节点图像的特征与第二节点对应的节点图像的特征之间的变化矩阵。计算机设备再对该变化矩阵进行分解得到旋转矩阵和位移矩阵，根据旋转矩阵得到图像帧的特征和节点图像的特征之间的空间角度差异量，根据位移矩阵得到图像帧的特征和节点图像的特征之间的空间位置差异量。计算机设备可再根据空间角度差异量确定当前运动的方向，根据空间位置差异量确定当前运动的距离，从而按照确定的方向移动确定的距离，运动至地图中的第二节点处。计算机设备可再按照相同的处理方式确定当前运动的距离和方向，依次从地图中的第二节点处按照趋向目标运动路径上运动，直至到达目标节点处。
在本实施例中，通过趋向目标运动路径包括的相邻节点所对应的节点图像的特征的空间状态差异量，逐步按照趋向目标运动路径在地图上从起始节点运动至目标节点，避免了在运动过程中发生偏差无法确定当前位置的问题， 保证了运动的准确性。
图9示出了一个实施例中在地图中选取趋向目标运动路径的示意图。参考图9，该示意图包括目标节点901、起始节点902以及趋向目标运动路径903。计算机设备在确定目标节点901即目标所在的位置以及起始节点902即本机所在位置后，以起始节点902为起点，以目标节点901为终点，在地图中选取目标运动路径903。
如图10所示，在一个具体的实施例中，运动控制方法包括以下步骤：
S1002，从按时序采集的图像帧中选取图像帧。
S1004，判断选取的图像帧的特征是否符合预设的节点图像的特征；若是，则跳转到步骤S1006；若否，则返回步骤S1002。
S1006，获取选取的图像帧为节点图像。
S1008，提取获取的节点图像的特征；获取地图中已有的节点对应的节点图像的特征；确定获取的特征与提取的特征之间的变化矩阵；根据节点与变化矩阵，确定获取的节点图像在地图中相应的节点，对应于确定的节点存储获取的节点图像的特征。
S1010，计算地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度；当地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度超过预设相似度阈值时，则根据获取的节点图像相应的节点，在地图中生成包括已有的节点的环形路径。
S1012，获取图像帧。
S1014，将图像帧输入卷积神经网络模型；获取卷积神经网络模型包括的多个网络层输出的特征图；将各特征图依次输入记忆神经网络模型，获取记忆神经网络模型输出的人脸检测结果。
S1016，判断人脸检测结果是否表示图像帧包括人脸图像；若是，则跳转到步骤S1018；若否，则返回步骤S1012。
S1018，提取人脸图像的人脸特征数据；根据人脸特征数据查询与人脸图像相匹配的预设人脸图像；根据预设人脸图像得到目标身份识别结果；确定 与目标身份识别结果相关联的服务类型。
S1020，确定人脸图像在地图中相应的目标节点。
S1022，提取图像帧的特征；获取地图包括的节点所对应的节点图像的特征；确定图像帧的特征与节点图像的特征之间的相似度；选取对应相似度最高的节点图像的特征所对应的节点，得到与图像帧相匹配的起始节点。
S1024，根据起始节点和目标节点，在地图包括的路径中选取趋向目标运动路径。
S1026，提取图像帧的特征；获取起始节点所对应的节点图像的特征；确定图像帧的特征和节点图像的特征之间的空间状态差异量；根据空间状态差异量进行运动。
S1028，依次获取趋向目标运动路径包括的各节点所对应的节点图像的特征；依次确定获取的对应相邻节点的节点图像的特征之间的空间状态差异量；根据依次确定的空间状态差异量进行运动。
S1030，提供与服务类型相应的服务触发入口。
在本实施例中，在获取到图像帧后，就可以自动地在检测到该图像帧包括人脸图像时，在地图中确定该人脸图像相应的目标节点，定位目标在地图中的位置，然后以该图像帧的特征与地图中各节点对应的节点图像的特征的匹配关系为依据，即可从地图中挑选与该图像帧匹配的起始节点，定位本机当前在地图中的位置，再根据当前节点和目标节点便可在地图包括的路径中选取趋向目标运动路径来运动。这样通过图像之间的特征匹配即可完成在地图中的定位，避免了通过传感信号定位引起的环境影响，提高了运动控制的准确性。
应该理解的是，虽然本申请各实施例中的各个步骤并不是必然按照步骤标号指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，各实施例中至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些子步骤或者 阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
如图11所示，在一个实施例中，提供了一种计算机设备1100。该计算机设备1100的内部结构可参照如图2所示的结构。下述的每个模块可全部或部分通过软件、硬件或其组合来实现。参照图11，该计算机设备1100包括：获取模块1101、确定模块1102、挑选模块1103、选取模块1104和运动模块1105。
获取模块1101，用于获取图像帧。
确定模块1102，用于当对图像帧进行人脸检测得到图像帧包括人脸图像时，确定人脸图像在地图中相应的目标节点。
挑选模块1103，用于从地图中挑选与图像帧匹配的起始节点；其中，图像帧的特征与起始节点对应的节点图像的特征相匹配。
选取模块1104，用于根据起始节点和目标节点，在地图包括的路径中选取趋向目标运动路径。
运动模块1105，用于按照选取的趋向目标运动路径运动。
上述计算机设备1100，在获取到图像帧后，就可以自动地在检测到该图像帧包括人脸图像时，在地图中确定该人脸图像相应的目标节点，定位目标在地图中的位置，然后以该图像帧的特征与地图中各节点对应的节点图像的特征的匹配关系为依据，即可从地图中挑选与该图像帧匹配的起始节点，定位本机当前在地图中的位置，再根据当前节点和目标节点便可在地图包括的路径中选取趋向目标运动路径来运动。这样通过图像之间的特征匹配即可完成在地图中的定位，避免了通过传感信号定位引起的环境影响，提高了运动控制的准确性。
如图12所示，在一个实施例中，计算机设备1100还包括：检测模块1106。
检测模块1106，用于将图像帧输入卷积神经网络模型；获取卷积神经网络模型包括的多个网络层输出的特征图；将各特征图依次输入记忆神经网络模型；获取记忆神经网络模型输出的图像帧是否包括人脸图像的结果。
在本实施例中，通过卷积神经网络模型的包括的多个网络层充分提取图 像特征，再将多层网络层提取的特征输入记忆神经网络模型综合处理，使得人脸检测更准确。
如图13所示，在一个实施例中，计算机设备1100还包括：识别模块1107和服务模块1108。
识别模块1107，用于提取人脸图像的人脸特征数据；根据人脸特征数据查询与人脸图像相匹配的预设人脸图像；根据预设人脸图像得到目标身份识别结果；确定与目标身份识别结果相关联的服务类型。
服务模块1108，用于提供与服务类型相应的服务触发入口。
在本实施例中，在对获取的图像检测到存在人脸时，对存在的人脸进行识别，在识别得到目标的身份并运动至目标后即可向该目标提供与该目标相关联的服务入口，极大地提高了服务提供的效率。
如图14所示，在一个实施例中，计算机设备1100还包括：地图构建模块1109。
地图构建模块1109，用于从按时序采集的图像帧中选取图像帧；判断选取的图像帧的特征是否符合预设的节点图像的特征；当选取的图像帧的特征符合节点图像的特征时，获取选取的图像帧为节点图像；确定获取的节点图像在地图中相应的节点；对应于确定的节点存储获取的节点图像的特征。
在本实施例中，通过自身采集图像帧，再对采集的图像帧进行处理即可自动进行地图构建，避免了需要大量具备专业绘图能力的工作人员人工对环境进行测绘，对工作人员能力要求高且劳动量大的问题，提高地图构建的效率。
在一个实施例中，地图构建模块1109还用于提取获取的节点图像的特征；获取地图中已有的节点对应的节点图像的特征；确定获取的特征与提取的特征之间的变化矩阵；根据节点与变化矩阵，确定获取的节点图像在地图中相应的节点。
在本实施例中，通过节点图像的特征之间的变化矩阵，得到当前获取的节点图像与在前已有的节点图像的转化关系，从而实现由在前的图像帧在地 图中的位置推测当前图像帧的在地图中的位置，实现实时定位。
在一个实施例中，地图构建模块1109还用于计算地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度；当地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度超过预设相似度阈值时，则根据获取的节点图像相应的节点，在地图中生成包括已有的节点的环形路径。
在本实施例中，以新增的节点图像的特征与已有的节点图像的特征的相似度作为依据进行闭环检测，在检测到有闭环时，在地图中生成环形路径，以进行后续的闭环优化，提高构建地图的准确性。
在一个实施例中，挑选模块1103还用于提取图像帧的特征；获取地图包括的节点所对应的节点图像的特征；确定图像帧的特征与节点图像的特征之间的相似度；选取对应相似度最高的节点图像的特征所对应的节点，得到与图像帧相匹配的起始节点。
在一个实施例中，运动模块1105还用于提取图像帧的特征；获取起始节点所对应的节点图像的特征；确定图像帧的特征和节点图像的特征之间的空间状态差异量；根据空间状态差异量进行运动。
在本实施例中，通过当前图像帧与地图包括的节点所对应的节点图像的特征相似匹配来定位当前在地图中的位置，使得自身定位结果更加准确。
在一个实施例中，运动模块1105还用于依次获取趋向目标运动路径包括的各节点所对应的节点图像的特征；依次确定获取的对应相邻节点的节点图像的特征之间的空间状态差异量；根据依次确定的空间状态差异量进行运动。
在本实施例中，通过当前获取到的图像帧与确定的初始节点对应的节点图像之间的空间状态差异量，以运动至地图中的初始节点处，从而按照选取的趋向目标运动路径向目标运动，保证了运动的准确性。
一种计算机可读存储介质，其上存储有计算机可读指令，该计算机可读指令被处理器执行时实现以下步骤：获取图像帧；当对图像帧进行人脸检测得到图像帧包括人脸图像时，确定人脸图像在地图中相应的目标节点；从地 图中挑选与图像帧匹配的起始节点；其中，图像帧的特征与起始节点对应的节点图像的特征相匹配；根据起始节点和目标节点，在地图包括的路径中选取趋向目标运动路径；按照选取的趋向目标运动路径运动。
上述计算机可读存储介质上存储的计算机可读指令在被执行时，在获取到图像帧后，就可以自动地在检测到该图像帧包括人脸图像时，在地图中确定该人脸图像相应的目标节点，定位目标在地图中的位置，然后以该图像帧的特征与地图中各节点对应的节点图像的特征的匹配关系为依据，即可从地图中挑选与该图像帧匹配的起始节点，定位本机当前在地图中的位置，再根据当前节点和目标节点便可在地图包括的路径中选取趋向目标运动路径来运动。这样通过图像之间的特征匹配即可完成在地图中的定位，避免了通过传感信号定位引起的环境影响，提高了运动控制的准确性。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：将图像帧输入卷积神经网络模型；获取卷积神经网络模型包括的多个网络层输出的特征图；将各特征图依次输入记忆神经网络模型；获取记忆神经网络模型输出的图像帧是否包括人脸图像的结果。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：提取人脸图像的人脸特征数据；根据人脸特征数据查询与人脸图像相匹配的预设人脸图像；根据预设人脸图像得到目标身份识别结果；确定与目标身份识别结果相关联的服务类型。计算机可读指令还使得处理器执行以下步骤：提供与服务类型相应的服务触发入口。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：从按时序采集的图像帧中选取图像帧；判断选取的图像帧的特征是否符合预设的节点图像的特征；当选取的图像帧的特征符合节点图像的特征时，获取选取的图像帧为节点图像；确定获取的节点图像在地图中相应的节点；对应于确定的节点存储获取的节点图像的特征。
在一个实施例中，确定获取的节点图像在地图中相应的节点，包括：提取获取的节点图像的特征；获取地图中已有的节点对应的节点图像的特征； 确定获取的特征与提取的特征之间的变化矩阵；根据节点与变化矩阵，确定获取的节点图像在地图中相应的节点。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：计算地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度；当地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度超过预设相似度阈值时，则根据获取的节点图像相应的节点，在地图中生成包括已有的节点的环形路径。
在一个实施例中，从地图中挑选与图像帧匹配的起始节点，包括：提取图像帧的特征；获取地图包括的节点所对应的节点图像的特征；确定图像帧的特征与节点图像的特征之间的相似度；选取对应相似度最高的节点图像的特征所对应的节点，得到与图像帧相匹配的起始节点。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：提取图像帧的特征；获取起始节点所对应的节点图像的特征；确定图像帧的特征和节点图像的特征之间的空间状态差异量；根据空间状态差异量进行运动。
在一个实施例中，按照选取的趋向目标运动路径运动，包括：依次获取趋向目标运动路径包括的各节点所对应的节点图像的特征；依次确定获取的对应相邻节点的节点图像的特征之间的空间状态差异量；根据依次确定的空间状态差异量进行运动。
一种计算机设备，包括存储器和处理器，存储器中储存有计算机可读指令，计算机可读指令被处理器执行时，使得处理器执行以下步骤：获取图像帧；当对图像帧进行人脸检测得到图像帧包括人脸图像时，确定人脸图像在地图中相应的目标节点；从地图中挑选与图像帧匹配的起始节点；其中，图像帧的特征与起始节点对应的节点图像的特征相匹配；根据起始节点和目标节点，在地图包括的路径中选取趋向目标运动路径；按照选取的趋向目标运动路径运动。
上述计算机设备，在获取到图像帧后，就可以自动地在检测到该图像帧包括人脸图像时，在地图中确定该人脸图像相应的目标节点，定位目标在地 图中的位置，然后以该图像帧的特征与地图中各节点对应的节点图像的特征的匹配关系为依据，即可从地图中挑选与该图像帧匹配的起始节点，定位本机当前在地图中的位置，再根据当前节点和目标节点便可在地图包括的路径中选取趋向目标运动路径来运动。这样通过图像之间的特征匹配即可完成在地图中的定位，避免了通过传感信号定位引起的环境影响，提高了运动控制的准确性。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：将图像帧输入卷积神经网络模型；获取卷积神经网络模型包括的多个网络层输出的特征图；将各特征图依次输入记忆神经网络模型；获取记忆神经网络模型输出的图像帧是否包括人脸图像的结果。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：提取人脸图像的人脸特征数据；根据人脸特征数据查询与人脸图像相匹配的预设人脸图像；根据预设人脸图像得到目标身份识别结果；确定与目标身份识别结果相关联的服务类型。计算机可读指令还使得处理器执行以下步骤：提供与服务类型相应的服务触发入口。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：从按时序采集的图像帧中选取图像帧；判断选取的图像帧的特征是否符合预设的节点图像的特征；当选取的图像帧的特征符合节点图像的特征时，获取选取的图像帧为节点图像；确定获取的节点图像在地图中相应的节点；对应于确定的节点存储获取的节点图像的特征。
在一个实施例中，确定获取的节点图像在地图中相应的节点，包括：提取获取的节点图像的特征；获取地图中已有的节点对应的节点图像的特征；确定获取的特征与提取的特征之间的变化矩阵；根据节点与变化矩阵，确定获取的节点图像在地图中相应的节点。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：计算地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度；当地图中已有的节点对应的节点图像的特征，与获取的节点图像的特 征之间的相似度超过预设相似度阈值时，则根据获取的节点图像相应的节点，在地图中生成包括已有的节点的环形路径。
在一个实施例中，从地图中挑选与图像帧匹配的起始节点，包括：提取图像帧的特征；获取地图包括的节点所对应的节点图像的特征；确定图像帧的特征与节点图像的特征之间的相似度；选取对应相似度最高的节点图像的特征所对应的节点，得到与图像帧相匹配的起始节点。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：提取图像帧的特征；获取起始节点所对应的节点图像的特征；确定图像帧的特征和节点图像的特征之间的空间状态差异量；根据空间状态差异量进行运动。
在一个实施例中，按照选取的趋向目标运动路径运动，包括：依次获取趋向目标运动路径包括的各节点所对应的节点图像的特征；依次确定获取的对应相邻节点的节点图像的特征之间的空间状态差异量；根据依次确定的空间状态差异量进行运动。
一种服务机器人，包括存储器和处理器，存储器中储存有计算机可读指令，计算机可读指令被处理器执行时，使得处理器执行以下步骤：获取图像帧；当对图像帧进行人脸检测得到图像帧包括人脸图像时，确定人脸图像在地图中相应的目标节点；从地图中挑选与图像帧匹配的起始节点；其中，图像帧的特征与起始节点对应的节点图像的特征相匹配；根据起始节点和目标节点，在地图包括的路径中选取趋向目标运动路径；按照选取的趋向目标运动路径运动。
上述服务机器人，在获取到图像帧后，就可以自动地在检测到该图像帧包括人脸图像时，在地图中确定该人脸图像相应的目标节点，定位目标在地图中的位置，然后以该图像帧的特征与地图中各节点对应的节点图像的特征的匹配关系为依据，即可从地图中挑选与该图像帧匹配的起始节点，定位本机当前在地图中的位置，再根据当前节点和目标节点便可在地图包括的路径中选取趋向目标运动路径来运动。这样通过图像之间的特征匹配即可完成在地图中的定位，避免了通过传感信号定位引起的环境影响，提高了运动控制 的准确性。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：将图像帧输入卷积神经网络模型；获取卷积神经网络模型包括的多个网络层输出的特征图；将各特征图依次输入记忆神经网络模型；获取记忆神经网络模型输出的图像帧是否包括人脸图像的结果。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：提取人脸图像的人脸特征数据；根据人脸特征数据查询与人脸图像相匹配的预设人脸图像；根据预设人脸图像得到目标身份识别结果；确定与目标身份识别结果相关联的服务类型。计算机可读指令还使得处理器执行以下步骤：提供与服务类型相应的服务触发入口。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：从按时序采集的图像帧中选取图像帧；判断选取的图像帧的特征是否符合预设的节点图像的特征；当选取的图像帧的特征符合节点图像的特征时，获取选取的图像帧为节点图像；确定获取的节点图像在地图中相应的节点；对应于确定的节点存储获取的节点图像的特征。
在一个实施例中，确定获取的节点图像在地图中相应的节点，包括：提取获取的节点图像的特征；获取地图中已有的节点对应的节点图像的特征；确定获取的特征与提取的特征之间的变化矩阵；根据节点与变化矩阵，确定获取的节点图像在地图中相应的节点。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：计算地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度；当地图中已有的节点对应的节点图像的特征，与获取的节点图像的特征之间的相似度超过预设相似度阈值时，则根据获取的节点图像相应的节点，在地图中生成包括已有的节点的环形路径。
在一个实施例中，从地图中挑选与图像帧匹配的起始节点，包括：提取图像帧的特征；获取地图包括的节点所对应的节点图像的特征；确定图像帧的特征与节点图像的特征之间的相似度；选取对应相似度最高的节点图像的 特征所对应的节点，得到与图像帧相匹配的起始节点。
在一个实施例中，计算机可读指令还使得处理器执行以下步骤：提取图像帧的特征；获取起始节点所对应的节点图像的特征；确定图像帧的特征和节点图像的特征之间的空间状态差异量；根据空间状态差异量进行运动。
在一个实施例中，按照选取的趋向目标运动路径运动，包括：依次获取趋向目标运动路径包括的各节点所对应的节点图像的特征；依次确定获取的对应相邻节点的节点图像的特征之间的空间状态差异量；根据依次确定的空间状态差异量进行运动。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一非易失性计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形 和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (20)

  1. 一种运动控制方法，包括：
    计算机设备获取图像帧；
    当对所述图像帧进行人脸检测得到所述图像帧包括人脸图像时，所述计算机设备确定所述人脸图像在地图中相应的目标节点；
    所述计算机设备从所述地图中挑选与所述图像帧匹配的起始节点；其中，所述图像帧的特征与所述起始节点对应的节点图像的特征相匹配；
    所述计算机设备根据所述起始节点和所述目标节点，在所述地图包括的路径中选取趋向目标运动路径；及
    所述计算机设备按照选取的所述趋向目标运动路径运动。
  2. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    所述计算机设备将所述图像帧输入卷积神经网络模型；
    所述计算机设备获取所述卷积神经网络模型包括的多个网络层输出的特征图；
    所述计算机设备将各所述特征图依次输入记忆神经网络模型；及
    所述计算机设备获取所述记忆神经网络模型输出的所述图像帧是否包括人脸图像的结果。
  3. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    所述计算机设备提取所述人脸图像的人脸特征数据；
    所述计算机设备根据所述人脸特征数据查询与所述人脸图像相匹配的预设人脸图像；
    所述计算机设备根据所述预设人脸图像得到目标身份识别结果；
    所述计算机设备确定与所述目标身份识别结果相关联的服务类型；及
    提供与所述服务类型相应的服务触发入口。
  4. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    所述计算机设备从按时序采集的图像帧中选取图像帧；
    所述计算机设备判断选取的图像帧的特征是否符合预设的节点图像的特 征；
    当选取的图像帧的特征符合所述节点图像的特征时，所述计算机设备获取选取的图像帧为节点图像；
    所述计算机设备确定获取的所述节点图像在地图中相应的节点；及
    所述计算机设备对应于确定的所述节点存储获取的所述节点图像的特征。
  5. 根据权利要求4所述的方法，其特征在于，所述计算机设备确定获取的所述节点图像在地图中相应的节点，包括：
    所述计算机设备提取获取的所述节点图像的特征；
    所述计算机设备获取地图中已有的节点对应的节点图像的特征；
    所述计算机设备确定获取的所述特征与提取的所述特征之间的变化矩阵；及
    所述计算机设备根据所述节点与所述变化矩阵，确定获取的所述节点图像在地图中相应的节点。
  6. 根据权利要求4所述的方法，其特征在于，所述方法还包括：
    所述计算机设备计算地图中已有的节点对应的节点图像的特征，与获取的所述节点图像的特征之间的相似度；及
    当地图中已有的节点对应的节点图像的特征，与获取的所述节点图像的特征之间的相似度超过预设相似度阈值时，所述计算机设备则根据获取的所述节点图像相应的节点，在所述地图中生成包括所述已有的节点的环形路径。
  7. 根据权利要求1所述的方法，其特征在于，所述计算机设备从所述地图中挑选与所述图像帧匹配的起始节点，包括：
    所述计算机设备提取所述图像帧的特征；
    所述计算机设备获取所述地图包括的节点所对应的节点图像的特征；
    所述计算机设备确定所述图像帧的特征与所述节点图像的特征之间的相似度；及
    所述计算机设备选取对应相似度最高的节点图像的特征所对应的节点， 得到与所述图像帧相匹配的起始节点。
  8. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    所述计算机设备提取所述图像帧的特征；
    所述计算机设备获取所述起始节点所对应的节点图像的特征；
    所述计算机设备确定所述图像帧的特征和所述节点图像的特征之间的空间状态差异量；及
    所述计算机设备根据所述空间状态差异量进行运动。
  9. 根据权利要求1所述的方法，其特征在于，所述计算机设备按照选取的所述趋向目标运动路径运动，包括：
    所述计算机设备依次获取所述趋向目标运动路径包括的各节点所对应的节点图像的特征；
    所述计算机设备依次确定获取的对应相邻节点的节点图像的特征之间的空间状态差异量；及
    所述计算机设备根据依次确定的所述空间状态差异量进行运动。
  10. 一个或多个存储有计算机可读指令的非易失性存储介质，所述计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器执行以下步骤：
    获取图像帧；
    当对所述图像帧进行人脸检测得到所述图像帧包括人脸图像时，确定所述人脸图像在地图中相应的目标节点；
    从所述地图中挑选与所述图像帧匹配的起始节点；其中，所述图像帧的特征与所述起始节点对应的节点图像的特征相匹配；
    根据所述起始节点和所述目标节点，在所述地图包括的路径中选取趋向目标运动路径；及
    按照选取的所述趋向目标运动路径运动。
  11. 根据权利要求10所述的存储介质，其特征在于，所述计算机可读指令被一个或多个处理器执行时，还使得一个或多个处理器执行以下步骤：
    提取所述人脸图像的人脸特征数据；
    根据所述人脸特征数据查询与所述人脸图像相匹配的预设人脸图像；
    根据所述预设人脸图像得到目标身份识别结果；
    确定与所述目标身份识别结果相关联的服务类型；及
    提供与所述服务类型相应的服务触发入口。
  12. 根据权利要求10所述的存储介质，其特征在于，所述计算机可读指令被一个或多个处理器执行时，还使得一个或多个处理器执行以下步骤：
    从按时序采集的图像帧中选取图像帧；
    判断选取的图像帧的特征是否符合预设的节点图像的特征；
    当选取的图像帧的特征符合所述节点图像的特征时，获取选取的图像帧为节点图像；
    确定获取的所述节点图像在地图中相应的节点；及
    对应于确定的所述节点存储获取的所述节点图像的特征。
  13. 根据权利要求10所述的存储介质，其特征在于，所述从所述地图中挑选与所述图像帧匹配的起始节点，包括：
    提取所述图像帧的特征；
    获取所述地图包括的节点所对应的节点图像的特征；
    确定所述图像帧的特征与所述节点图像的特征之间的相似度；及
    选取对应相似度最高的节点图像的特征所对应的节点，得到与所述图像帧相匹配的起始节点。
  14. 根据权利要求10所述的存储介质，其特征在于，所述计算机可读指令被一个或多个处理器执行时，还使得一个或多个处理器执行以下步骤：
    提取所述图像帧的特征；
    获取所述起始节点所对应的节点图像的特征；
    确定所述图像帧的特征和所述节点图像的特征之间的空间状态差异量；及
    根据所述空间状态差异量进行运动。
  15. 一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行以下步骤：
    获取图像帧；
    当对所述图像帧进行人脸检测得到所述图像帧包括人脸图像时，确定所述人脸图像在地图中相应的目标节点；
    从所述地图中挑选与所述图像帧匹配的起始节点；其中，所述图像帧的特征与所述起始节点对应的节点图像的特征相匹配；
    根据所述起始节点和所述目标节点，在所述地图包括的路径中选取趋向目标运动路径；及
    按照选取的所述趋向目标运动路径运动。
  16. 根据权利要求15所述的计算机设备，其特征在于，所述计算机可读指令被所述处理器执行时，还使得所述处理器执行以下步骤：
    提取所述人脸图像的人脸特征数据；
    根据所述人脸特征数据查询与所述人脸图像相匹配的预设人脸图像；
    根据所述预设人脸图像得到目标身份识别结果；
    确定与所述目标身份识别结果相关联的服务类型；及
    提供与所述服务类型相应的服务触发入口。
  17. 根据权利要求15所述的计算机设备，其特征在于，所述计算机可读指令被所述处理器执行时，还使得所述处理器执行以下步骤：
    从按时序采集的图像帧中选取图像帧；
    判断选取的图像帧的特征是否符合预设的节点图像的特征；
    当选取的图像帧的特征符合所述节点图像的特征时，获取选取的图像帧为节点图像；
    确定获取的所述节点图像在地图中相应的节点；及
    对应于确定的所述节点存储获取的所述节点图像的特征。
  18. 根据权利要求15所述的计算机设备，其特征在于，所述从所述地图 中挑选与所述图像帧匹配的起始节点，包括：
    提取所述图像帧的特征；
    获取所述地图包括的节点所对应的节点图像的特征；
    确定所述图像帧的特征与所述节点图像的特征之间的相似度；及
    选取对应相似度最高的节点图像的特征所对应的节点，得到与所述图像帧相匹配的起始节点。
  19. 根据权利要求10所述的计算机设备，其特征在于，所述计算机可读指令被所述处理器执行时，还使得所述处理器执行以下步骤：
    提取所述图像帧的特征；
    获取所述起始节点所对应的节点图像的特征；
    确定所述图像帧的特征和所述节点图像的特征之间的空间状态差异量；及
    根据所述空间状态差异量进行运动。
  20. 一种服务机器人，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述计算机可读指令被所述处理器执行时，使得所述处理器执行以下步骤：
    获取图像帧；
    当对所述图像帧进行人脸检测得到所述图像帧包括人脸图像时，确定所述人脸图像在地图中相应的目标节点；
    从所述地图中挑选与所述图像帧匹配的起始节点；其中，所述图像帧的特征与所述起始节点对应的节点图像的特征相匹配；
    根据所述起始节点和所述目标节点，在所述地图包括的路径中选取趋向目标运动路径；及
    按照选取的所述趋向目标运动路径运动。
PCT/CN2018/085065 2017-05-22 2018-04-28 运动控制方法、存储介质、计算机设备和服务机器人 WO2018214706A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710365516.XA CN107341442B (zh) 2017-05-22 2017-05-22 运动控制方法、装置、计算机设备和服务机器人
CN201710365516.X 2017-05-22

Publications (1)

Publication Number Publication Date
WO2018214706A1 true WO2018214706A1 (zh) 2018-11-29

Family

ID=60221306

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/085065 WO2018214706A1 (zh) 2017-05-22 2018-04-28 运动控制方法、存储介质、计算机设备和服务机器人

Country Status (2)

Country Link
CN (1) CN107341442B (zh)
WO (1) WO2018214706A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111144275A (zh) * 2019-12-24 2020-05-12 中石化第十建设有限公司 一种基于人脸识别的跑步智能测试系统及方法
CN111241943A (zh) * 2019-12-31 2020-06-05 浙江大学 自动驾驶场景下基于背景目标检测与三元组损失的场景识别与回环检测方法
CN111506104A (zh) * 2020-04-03 2020-08-07 北京邮电大学 一种规划无人机位置的方法及装置
CN111815738A (zh) * 2020-06-15 2020-10-23 北京沃东天骏信息技术有限公司 一种构建地图的方法和装置
CN112464989A (zh) * 2020-11-02 2021-03-09 北京科技大学 一种基于目标检测网络的闭环检测方法
CN112528728A (zh) * 2020-10-16 2021-03-19 深圳市银星智能科技股份有限公司 一种用于视觉导航的图像处理方法、装置和移动机器人
US11466988B2 (en) * 2018-12-13 2022-10-11 Goertek Inc. Method and device for extracting key frames in simultaneous localization and mapping and smart device

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107341442B (zh) * 2017-05-22 2023-06-06 腾讯科技（上海）有限公司 运动控制方法、装置、计算机设备和服务机器人
CN108236777A (zh) * 2018-01-08 2018-07-03 深圳市易成自动驾驶技术有限公司 拾球方法、拾球车及计算机可读存储介质
KR20200010640A (ko) * 2018-06-27 2020-01-31 삼성전자주식회사 모션 인식 모델을 이용한 자체 운동 추정 장치 및 방법, 모션 인식 모델 트레이닝 장치 및 방법
CN110794951A (zh) * 2018-08-01 2020-02-14 北京京东尚科信息技术有限公司 一种基于用户动作确定购物指令的方法和装置
CN109389156B (zh) * 2018-09-11 2022-05-03 深圳大学 一种图像定位模型的训练方法、装置及图像定位方法
CN113343739B (zh) * 2020-03-02 2022-07-22 杭州萤石软件有限公司 可移动设备的重定位方法和可移动设备
CN112914601B (zh) * 2021-01-19 2024-04-02 深圳市德力凯医疗设备股份有限公司 一种机械臂的避障方法、装置、存储介质及超声设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102411368A (zh) * 2011-07-22 2012-04-11 北京大学 机器人的主动视觉人脸跟踪方法和跟踪系统
CN104236548A (zh) * 2014-09-12 2014-12-24 清华大学 一种微型无人机室内自主导航方法
US20150227775A1 (en) * 2012-09-11 2015-08-13 Southwest Research Institute 3-D Imaging Sensor Based Location Estimation
CN105911992A (zh) * 2016-06-14 2016-08-31 广东技术师范学院 一种移动机器人的自动规划路径方法及移动机器人
CN107341442A (zh) * 2017-05-22 2017-11-10 腾讯科技（上海）有限公司 运动控制方法、装置、计算机设备和服务机器人

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4282662B2 (ja) * 2004-12-14 2009-06-24 本田技研工業株式会社 自律移動ロボットの移動経路生成装置
US20140031980A1 (en) * 2011-11-11 2014-01-30 Jens-Steffen Gutmann Systems and methods for extending slam to multiple regions
US10466056B2 (en) * 2014-04-25 2019-11-05 Samsung Electronics Co., Ltd. Trajectory matching using ambient signals
US10068373B2 (en) * 2014-07-01 2018-09-04 Samsung Electronics Co., Ltd. Electronic device for providing map information
JP2015180974A (ja) * 2015-07-17 2015-10-15 株式会社ナビタイムジャパン 階層別地図データを備えた情報処理システム、情報処理プログラム、情報処理装置、および、情報処理方法
CN106125730B (zh) * 2016-07-10 2019-04-30 北京工业大学 一种基于鼠脑海马空间细胞的机器人导航地图构建方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102411368A (zh) * 2011-07-22 2012-04-11 北京大学 机器人的主动视觉人脸跟踪方法和跟踪系统
US20150227775A1 (en) * 2012-09-11 2015-08-13 Southwest Research Institute 3-D Imaging Sensor Based Location Estimation
CN104236548A (zh) * 2014-09-12 2014-12-24 清华大学 一种微型无人机室内自主导航方法
CN105911992A (zh) * 2016-06-14 2016-08-31 广东技术师范学院 一种移动机器人的自动规划路径方法及移动机器人
CN107341442A (zh) * 2017-05-22 2017-11-10 腾讯科技（上海）有限公司 运动控制方法、装置、计算机设备和服务机器人

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11466988B2 (en) * 2018-12-13 2022-10-11 Goertek Inc. Method and device for extracting key frames in simultaneous localization and mapping and smart device
CN111144275A (zh) * 2019-12-24 2020-05-12 中石化第十建设有限公司 一种基于人脸识别的跑步智能测试系统及方法
CN111241943A (zh) * 2019-12-31 2020-06-05 浙江大学 自动驾驶场景下基于背景目标检测与三元组损失的场景识别与回环检测方法
CN111241943B (zh) * 2019-12-31 2022-06-21 浙江大学 基于背景目标与三元组损失的场景识别与回环检测方法
CN111506104A (zh) * 2020-04-03 2020-08-07 北京邮电大学 一种规划无人机位置的方法及装置
CN111506104B (zh) * 2020-04-03 2021-10-01 北京邮电大学 一种规划无人机位置的方法及装置
CN111815738A (zh) * 2020-06-15 2020-10-23 北京沃东天骏信息技术有限公司 一种构建地图的方法和装置
CN111815738B (zh) * 2020-06-15 2024-01-12 北京京东乾石科技有限公司 一种构建地图的方法和装置
CN112528728A (zh) * 2020-10-16 2021-03-19 深圳市银星智能科技股份有限公司 一种用于视觉导航的图像处理方法、装置和移动机器人
CN112528728B (zh) * 2020-10-16 2024-03-29 深圳银星智能集团股份有限公司 一种用于视觉导航的图像处理方法、装置和移动机器人
CN112464989A (zh) * 2020-11-02 2021-03-09 北京科技大学 一种基于目标检测网络的闭环检测方法
CN112464989B (zh) * 2020-11-02 2024-02-20 北京科技大学 一种基于目标检测网络的闭环检测方法

Also Published As

Publication number Publication date
CN107341442A (zh) 2017-11-10
CN107341442B (zh) 2023-06-06

Similar Documents

Publication Publication Date Title
WO2018214706A1 (zh) 运动控制方法、存储介质、计算机设备和服务机器人
US10810734B2 (en) Computer aided rebar measurement and inspection system
WO2021196294A1 (zh) 一种跨视频人员定位追踪方法、系统及设备
Walch et al. Image-based localization using lstms for structured feature correlation
CN109919977B (zh) 一种基于时间特征的视频运动人物跟踪与身份识别方法
CN104142939B (zh) 一种基于运动特征信息匹配特征码的方法与设备
JP2021531592A (ja) 追跡ターゲットの位置決め方法、装置、機器及び記憶媒体
WO2020052476A1 (zh) 特征点定位方法、存储介质和计算机设备
CN109934847B (zh) 弱纹理三维物体姿态估计的方法和装置
WO2020134528A1 (zh) 目标检测方法及相关产品
WO2019019935A1 (zh) 交互方法、交互终端、存储介质和计算机设备
JP2022542204A (ja) 画像座標系の変換方法並びにその、装置、機器およびコンピュータプログラム
JP2009157767A (ja) 顔画像認識装置、顔画像認識方法、顔画像認識プログラムおよびそのプログラムを記録した記録媒体
TWI745818B (zh) 視覺定位方法、電子設備及電腦可讀儲存介質
KR102464271B1 (ko) 포즈 획득 방법, 장치, 전자 기기, 저장 매체 및 프로그램
CN102100063A (zh) 视频处理
US20210209849A1 (en) Multiple maps for 3d object scanning and reconstruction
Zheng et al. Buildingfusion: Semantic-aware structural building-scale 3d reconstruction
CN114266823A (zh) 一种结合SuperPoint网络特征提取的单目SLAM方法
CN114399515A (zh) 基于语言描述的类级别目标物6d位姿获取方法及存储介质
CN116563352A (zh) 融合深度视觉信息的单线激光雷达回环检测方法及系统
WO2023178951A1 (zh) 图像分析方法、模型的训练方法、装置、设备、介质及程序
Sinha et al. Image retrieval using landmark indexing for indoor navigation
CN110849380A (zh) 一种基于协同vslam的地图对齐方法及系统
JP2015184743A (ja) 画像処理装置および物体認識方法

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: 18805492

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: 18805492

Country of ref document: EP

Kind code of ref document: A1