CN-121541648-B - Method, device, equipment and medium for foot robot target following

CN121541648BCN 121541648 BCN121541648 BCN 121541648BCN-121541648-B

Abstract

The invention provides a method, a device, equipment and a medium for following a target of a foot robot, which comprise the steps of carrying out boundary frame representation on a following target to obtain a real-time boundary frame, matching a central point coordinate of the real-time detection frame with a central point coordinate of the real-time boundary frame to obtain a following point coordinate, carrying out coordinate transformation on the following point coordinate to obtain a point cloud central coordinate, taking each point in the point cloud in a passable area as a vertex, setting the weight of edges between the vertices to construct a passing point cloud graph, taking a point corresponding to the point cloud central coordinate as a source point, presetting a background point as a sink point, identifying the minimum cut from the passing point cloud graph to obtain a point cloud cluster, and controlling the foot robot to carry out real-time following on the following target through the point cloud cluster corresponding to the following target. Therefore, the robot can accurately lock the following target in a dynamically changing environment, can avoid obstacles autonomously, and ensures the stability and safety of the following process.

Inventors

Request for anonymity
Request for anonymity

Assignees

杭州云深处科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260112

Claims (10)

1. A method of foot robot target following comprising: carrying out boundary frame representation on a following target in a last video frame acquired by the foot robot to obtain a real-time boundary frame of the following target; Matching the center point coordinates of the real-time detection frame of the current video frame with the center point coordinates of the real-time boundary frame to obtain following point coordinates corresponding to the current video frame; carrying out coordinate transformation on the following point coordinates so as to convert the following point coordinates into a radar coordinate system, and obtaining point cloud center coordinates of the following point coordinates corresponding to the radar coordinate system; taking each point in the point cloud in the passable area as a vertex, setting the weight of the edges between the vertices, and constructing to obtain a passing point cloud image; Taking a point corresponding to the point cloud center coordinates as a source point, presetting a background point as a sink point, and identifying a minimum cut from the passing point cloud images so as to carry out target division on the passing point cloud images and obtain a point cloud cluster corresponding to the following target; And controlling the foot robot to follow the following target in real time through the point cloud cluster corresponding to the following target.
2. The method of claim 1, wherein matching the center point coordinates of the real-time detection frame of the current video frame with the center point coordinates of the real-time bounding frame to obtain following point coordinates corresponding to the current video frame comprises: Matching the center point coordinates of each real-time detection frame in the current video frame with the center point coordinates of the real-time boundary frame, and calculating the similarity cost between the real-time boundary frame and each real-time detection frame; Constructing a cost matrix based on the similarity cost between the real-time boundary frame and each real-time detection frame, and solving the cost matrix through a Hungary algorithm to obtain the optimal matching between the real-time boundary frame and the real-time detection frames; And determining the center point coordinate of one real-time detection frame successfully matched with the real-time boundary frame in the optimal matching result as the following point coordinate corresponding to the current video frame.
3. The method of claim 1, wherein obtaining a passable area for the foot robot under the current video frame comprises: determining a local accumulated point cloud of the foot robot in a current field of view based on pose transformation information of the foot robot at a current position point and a registration point cloud of the foot robot corresponding to a body coordinate system; And carrying out passable judgment on the foot robot through the local accumulated point cloud to obtain a passable area corresponding to the foot robot under the current video frame.
4. The method of claim 1, wherein identifying a minimum cut from the passing point cloud graph with a point corresponding to the point cloud center coordinates as a source point and a preset background point as a sink point to perform target division on the passing point cloud graph to obtain the point cloud cluster corresponding to the following target, comprises: Taking a point corresponding to the point cloud center coordinates as a source point, connecting the source point with all foreground points in the passing point cloud image, taking a preset background point as a sink point, and connecting the sink point with all background points in the passing point cloud image to obtain a directed image corresponding to the passing point cloud image; Solving the maximum flow from the source point to the sink point in the directed graph through a maximum flow algorithm to obtain a minimum cut; and carrying out target division on the passing point cloud image through the minimum cut to obtain the point cloud cluster corresponding to the following target.
5. The method of claim 1, wherein controlling the foot robot to follow the following target in real time through the point cloud cluster corresponding to the following target comprises: Calculating real-time position coordinates and gesture information of the following target in a three-dimensional space based on the point cloud cluster, wherein the real-time position coordinates are determined by the average value of three-dimensional coordinates of all points in the point cloud cluster, and the gesture information is used for describing an orientation vector obtained according to the principal component analysis result of the point cloud cluster; And taking the real-time position coordinates and the gesture information as a following reference standard of the foot robot, and adjusting the motion parameters of each joint of the foot robot in real time through a pre-generated target following path so as to drive the foot robot to follow the following target in real time.
6. The method according to claim 1, wherein the step of performing bounding box representation on the following target in the last video frame acquired by the foot robot to obtain a real-time bounding box of the following target comprises: Performing feature extraction and target identification on a previous video frame through a preset target detection algorithm to obtain pixel level position information of a following target in the previous video frame; and generating a boundary bounding box capable of completely wrapping the following target based on the pixel-level position information so as to obtain a real-time boundary box of the following target.
7. The method as recited in claim 1, further comprising: In the process of controlling the foot robot to follow the following target in real time, displaying a real-time boundary frame of the following target and real-time pose information of the following target in real time through a handle end; When the real-time boundary box is detected to exceed the preset visual field safety range, a corresponding early warning prompt is generated, and risk early warning is carried out on an operator of the foot robot in a mode of audible and visual warning or early warning signal sending to a control terminal.
8. A foot robot target following apparatus, comprising: The representation module is used for carrying out boundary frame representation on a following target in a previous video frame acquired by the foot robot to obtain a real-time boundary frame of the following target; The matching module is used for matching the center point coordinates of the real-time detection frame of the current video frame with the center point coordinates of the real-time boundary frame to obtain following point coordinates corresponding to the current video frame; the conversion module is used for carrying out coordinate transformation on the following point coordinates so as to convert the following point coordinates into a radar coordinate system, and obtaining point cloud center coordinates of the following point coordinates corresponding to the radar coordinate system; The construction module is used for acquiring a passable area corresponding to the foot robot under the current video frame, taking each point in the point cloud in the passable area as a vertex, setting the weight of the edges between the vertices, and constructing to obtain a passing point cloud image; The dividing module is used for taking a point corresponding to the point cloud center coordinates as a source point, presetting a background point as a sink point, identifying a minimum cut from the passing point cloud image, and carrying out target division on the passing point cloud image to obtain a point cloud cluster corresponding to the following target; and the control module is used for controlling the foot robot to follow the following target in real time through the point cloud cluster corresponding to the following target.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the processor, when executing the computer program, implements a method for target following by a foot robot according to any one of claims 1 to 7.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements a method of object following of a foot robot according to any one of claims 1-7.

Description

Method, device, equipment and medium for foot robot target following Technical Field Embodiments of the present disclosure relate to the field of artificial intelligence, and in particular, to a method, apparatus, device, and medium suitable for target following of a foot robot. Background The foot robot is a robot which realizes movement and interaction functions through a multi-foot limb structure, is generally provided with a plurality of leg execution mechanisms which can be independently driven, and can complete actions such as walking, crossing, obstacle avoidance and the like in a complex terrain environment. In the related art, when a foot robot performs a target following task, the foot robot mainly depends on the acquired environment and target information and drives the legs to move through a preset control algorithm so as to track the target. However, the following robustness and adaptability of the prior art scheme in a complex dynamic environment are insufficient, so that a larger position deviation and even a following failure occur in the following process. Disclosure of Invention Embodiments described herein provide a method, apparatus, device, and medium for foot robot target following that overcomes the above-described problems. In a first aspect, according to the present disclosure, there is provided a method of foot robot target following, comprising: carrying out boundary frame representation on a following target in a last video frame acquired by the foot robot to obtain a real-time boundary frame of the following target; Matching the center point coordinates of the real-time detection frame of the current video frame with the center point coordinates of the real-time boundary frame to obtain following point coordinates corresponding to the current video frame; carrying out coordinate transformation on the following point coordinates so as to convert the following point coordinates into a radar coordinate system, and obtaining point cloud center coordinates of the following point coordinates corresponding to the radar coordinate system; taking each point in the point cloud in the passable area as a vertex, setting the weight of the edges between the vertices, and constructing to obtain a passing point cloud image; Taking a point corresponding to the point cloud center coordinates as a source point, presetting a background point as a sink point, and identifying a minimum cut from the passing point cloud images so as to carry out target division on the passing point cloud images and obtain a point cloud cluster corresponding to the following target; And controlling the foot robot to follow the following target in real time through the point cloud cluster corresponding to the following target. In a second aspect, according to the present disclosure, there is provided an apparatus for foot robot target following, comprising: The representation module is used for carrying out boundary frame representation on a following target in a previous video frame acquired by the foot robot to obtain a real-time boundary frame of the following target; The matching module is used for matching the center point coordinates of the real-time detection frame of the current video frame with the center point coordinates of the real-time boundary frame to obtain following point coordinates corresponding to the current video frame; the conversion module is used for carrying out coordinate transformation on the following point coordinates so as to convert the following point coordinates into a radar coordinate system, and obtaining point cloud center coordinates of the following point coordinates corresponding to the radar coordinate system; The construction module is used for acquiring a passable area corresponding to the foot robot under the current video frame, taking each point in the point cloud in the passable area as a vertex, setting the weight of the edges between the vertices, and constructing to obtain a passing point cloud image; The dividing module is used for taking a point corresponding to the point cloud center coordinates as a source point, presetting a background point as a sink point, identifying a minimum cut from the passing point cloud image, and carrying out target division on the passing point cloud image to obtain a point cloud cluster corresponding to the following target; and the control module is used for controlling the foot robot to follow the following target in real time through the point cloud cluster corresponding to the following target. In a third aspect, a computer device is provided, comprising a memory in which a computer program is stored, and a processor which, when executing the computer program, performs the steps of the method of foot robot target following as in any of the embodiments above. In a fourth aspect, a computer readable storage medium is provided, on which a computer program is stored which, when executed by a processor, implements the steps of a method for