CN-121979209-A - Robot obstacle avoidance control system and method based on visual recognition

CN121979209ACN 121979209 ACN121979209 ACN 121979209ACN-121979209-A

Abstract

The invention provides a robot obstacle avoidance control system and method based on visual recognition, which relates to the technical field of image recognition control, and constructs a space-time characteristic point sequence by collecting image data in front of a robot, calculates obstacle index data, divides obstacles into obstacle types, respectively constructs a prediction model, meanwhile, the obstacles are grouped based on obstacle indexes, a space association diagram is constructed, the obstacle states are estimated in a layered mode, finally, an obstacle avoidance priority decision tree is constructed based on a prediction model and state estimation results, and the movement direction and speed of the robot are controlled according to the decision tree output. Therefore, the problems that the obstacle state in the complex shielding scene is difficult to accurately predict and the obstacle avoidance strategy is inflexible can be solved, and the obstacle avoidance precision and safety of the robot in the complex environment are improved.

Inventors

QIU XUDONG
XU WEITING
WANG WENWEI

Assignees

浙江科聪控制技术有限公司

Dates

Publication Date: 20260505
Application Date: 20260116

Claims (10)

1. The robot obstacle avoidance control method based on visual recognition is characterized by comprising the following steps of: acquiring a multi-scale edge feature map of a front image of the robot, constructing a space-time feature point sequence on the multi-scale edge feature map, and calculating barrier index data; Dividing an obstacle into a stable movement obstacle and an unstable movement obstacle based on the obstacle index data, establishing a movement track prediction model for the stable movement obstacle, and establishing a multi-hypothesis state estimation model for the unstable movement obstacle; Grouping the obstacles based on the obstacle index data, establishing a space correlation diagram among the obstacles, carrying out layered estimation on the state of the obstacles based on the space correlation diagram, and carrying out track tracking on the obstacles with different shielding levels by combining a prediction model; And constructing a decision tree of obstacle avoidance priority based on the motion trail prediction model, the multi-hypothesis state estimation model and the estimation result of the obstacle state, and controlling the motion direction and the speed of the robot according to the output of the decision tree.
2. The vision recognition-based robot obstacle avoidance control method according to claim 1, wherein a motion trajectory prediction model based on kalman filtering is adopted based on the stable motion obstacle, the position, the speed and the acceleration are used as state variables, the trajectory prediction is performed by using observed data in the past time, and the noise covariance matrix is dynamically adjusted according to reliability of feature point tracking and feature matching.
3. The vision recognition-based robot obstacle avoidance control method according to claim 2, wherein based on the multi-hypothesis state estimation model, probability weights of modes are dynamically adjusted by using an interactive multi-model algorithm through hypotheses of three basic motion modes of uniform motion, uniform acceleration motion and steering motion, each mode corresponding to an independent kalman filter.
4. The vision-recognition-based robot obstacle avoidance control method of claim 1, wherein the obstacle indicator data comprises a motion continuity indicator, an occlusion status indicator, and an occlusion depth indicator.
5. The vision recognition-based robot obstacle avoidance control method according to claim 4, wherein the obstacles are grouped based on the obstacle status index, whether the obstacle relationship exists is judged, the obstacles with the obstacle relationship are divided into obstacle groups, a spatial association graph is constructed in each obstacle group, the obstacle attribute among the obstacles is recorded, and the abnormal relationship is eliminated through time sequence consistency check.
6. The vision recognition-based robot obstacle avoidance control method according to claim 5, wherein multiple shielding conditions are analyzed based on the shielding depth index, when depth jump in an area is remarkable, and a characteristic point presents a multi-level vanishing-reappearance mode, the multiple shielding area is determined, a processing mode is activated, and a shielding level of an obstacle is determined by establishing a directed shielding relation diagram and combining a shielding area proportion, a depth jump amplitude and shielding duration.
7. The vision recognition-based robot obstacle avoidance control method according to claim 6, wherein a differential tracking strategy is adopted based on different shielding levels, wherein the foreground layer uses complete feature point tracking, the middle layer processes feature point disappearance and position correction through a prediction-verification framework, and the background layer relies on motion model prediction and utilizes historical track to match feature points.
8. The vision recognition-based robot obstacle avoidance control method according to claim 7, wherein tracking results of the differential tracking strategy are input into an extended kalman filter frame, and the obstacle states are optimally estimated in combination with interlayer constraints, so that a complete scene structure is reconstructed, and the geometric outline of the blocked area is restored.
9. The vision recognition-based robot obstacle avoidance control method of claim 8, wherein an obstacle avoidance priority decision tree is constructed based on the motion trajectory prediction model, the multi-hypothesis state estimation model, and the estimation result of the obstacle state; The obstacle avoidance priority decision tree divides the surrounding environment into sectors, calculates a risk score through the movement type, the shielding state and the confidence of the obstacle, dynamically adjusts the weight of the sectors, and selects the direction with the minimum weight to avoid the obstacle.
10. A robot obstacle avoidance control system based on visual recognition, which controls robot obstacle avoidance using the robot obstacle avoidance control method based on visual recognition according to any one of claims 1 to 9, comprising: the acquisition module acquires front image data of the robot, constructs a space-time characteristic point sequence and calculates barrier index data; the prediction module is used for dividing the obstacle into obstacle types based on the obstacle index data and respectively constructing prediction models; The classification module is used for grouping the obstacles based on the obstacle index data, establishing a space association diagram among the obstacles, carrying out layered estimation on the state of the obstacles, and carrying out track tracking on the obstacles with different shielding levels by combining a prediction model; And the control module is used for constructing a decision tree of obstacle avoidance priority based on the prediction model and the estimation result of the obstacle state, and controlling the movement direction and speed of the robot according to the output of the decision tree.

Description

Robot obstacle avoidance control system and method based on visual recognition Technical Field The invention relates to the technical field of image recognition control, in particular to a robot obstacle avoidance control system and method based on visual recognition. Background In recent years, with the rapid development of artificial intelligence and robot technology, a robot obstacle avoidance system based on computer vision is becoming an important research direction in the field of intelligent robots. The robot obstacle avoidance control technology is widely applied to scenes such as industrial automation, unmanned, service robots and the like, and is characterized in that environmental information is sensed through a sensor, and effective avoidance of dynamic or static obstacles is realized by combining a motion planning algorithm. In the visual recognition technology, the combination of the camera and the depth sensor provides a robot with rich environment sensing capability, particularly the appearance of the RGB-D sensor, so that the robot can acquire the color information and the depth information of an image at the same time, and an important basis is provided for multi-scale feature extraction, obstacle detection and track prediction. Currently, an obstacle avoidance control method based on visual recognition generally relies on the collaborative work of sub-modules such as feature extraction and tracking, motion track prediction, obstacle state estimation, obstacle avoidance strategy optimization, and the like. However, with the increasing complexity of the environment, such as multi-obstacle shielding, complex movement of dynamic obstacles, discontinuous depth information, etc., the conventional obstacle avoidance method faces serious challenges in terms of accuracy, real-time and stability. In the prior art, most robotic obstacle avoidance methods rely on feature extraction and analysis of a single scale or single information source (e.g., based on RGB images or depth maps only). The method is often insufficient in processing complex dynamic environments, firstly, in terms of multi-scale feature extraction, due to the lack of comprehensive analysis on multi-scale pyramid image features, the existing method is difficult to capture detailed information of obstacles under different scales, particularly in dynamic occlusion areas, the robustness of feature extraction is poor, secondly, in terms of occlusion state and hierarchy analysis of the obstacles, the existing technology is dependent on static occlusion models, and is difficult to cope with complex scenes of dynamic changes and multi-level occlusion of the obstacles, and in terms of track prediction of the dynamic obstacles, the existing technology usually adopts a single model (such as simple Kalman filtering or particle filtering) to perform track estimation, which shows obvious limitations in processing obstacles of unstable motion or steering motion. Especially under the condition that a shielding area is serious or a motion state is complex, the reliability of motion prediction and obstacle avoidance decision of the obstacle by the existing method is obviously reduced, so that the robot path planning efficiency is low, and even the risk of obstacle avoidance failure is caused. Disclosure of Invention The present invention has been made to solve the above-mentioned technical problems. The invention provides a robot obstacle avoidance control system and method based on visual identification, which can solve the problems that multi-level shielding and dynamic obstacle movement in a complex environment are difficult to effectively solve due to the defects in multi-scale feature extraction, dynamic shielding treatment and track prediction methods to a certain extent. According to one aspect of the present invention, there is provided a robot obstacle avoidance control method based on visual recognition, including: acquiring a multi-scale edge feature map of a front image of the robot, constructing a space-time feature point sequence on the multi-scale edge feature map, and calculating barrier index data; Dividing an obstacle into a stable movement obstacle and an unstable movement obstacle based on the obstacle index data, establishing a movement track prediction model for the stable movement obstacle, and establishing a multi-hypothesis state estimation model for the unstable movement obstacle; Grouping the obstacles based on the obstacle index data, establishing a space correlation diagram among the obstacles, carrying out layered estimation on the state of the obstacles based on the space correlation diagram, and carrying out track tracking on the obstacles with different shielding levels by combining a prediction model; And constructing a decision tree of obstacle avoidance priority based on the motion trail prediction model, the multi-hypothesis state estimation model and the estimation result of the obstacle state, and controlling the