CN-122024199-A - Dynamic obstacle scene vision passable area identification method

CN122024199ACN 122024199 ACN122024199 ACN 122024199ACN-122024199-A

Abstract

The invention relates to the field of robot vision perception, in particular to a method for identifying a vision passable area of a dynamic obstacle scene, which comprises the steps of determining a supporting state, an occupying state and a state to be verified of a local space unit according to multi-view time sequence images and pose increment data, and constructing a three-state space distribution diagram; the method comprises the steps of authenticating a passing unit based on a robot body envelope and a dynamic obstacle occupation time window to generate a candidate passing corridor, executing active supplementary observation and carrying out projection consistency verification when a local space unit in a state to be verified exists, updating a three-state space distribution diagram according to an authentication result and a verification result, and outputting an instant passing corridor, a time window passing corridor or an non-passing area. The invention reduces misjudgment caused by shielding and dynamic interference.

Inventors

DING CHAO
ZHANG JUNLI
TIAN JIASEN
CHEN BO
QU MIN
LI BO
TAN XIAO
Han Silei

Assignees

湘西民族职业技术学院

Dates

Publication Date: 20260512
Application Date: 20260403

Claims (10)

1. A method for identifying a visually passable region of a dynamic barrier scene, the method comprising: According to the multi-view time sequence image and the pose incremental data, determining the supporting state, the occupying state and the state to be verified of the local space unit, and constructing a three-state space distribution diagram; Based on the robot body envelope and the dynamic obstacle occupation time window, carrying out support continuity, traffic clearance, connectivity and time window authentication on traffic units in the three-state space distribution diagram, and generating candidate traffic corridor; Under the condition that a local space unit in a state to be verified exists in a traffic unit corresponding to the candidate traffic corridor, executing an active supplementary observation action, acquiring a supplementary observation image, and carrying out projection consistency verification on the candidate traffic corridor based on the supplementary observation image; and updating the three-state space distribution diagram according to the passing unit authentication result and the projection consistency check result, and outputting an instant passable corridor, a time window passable corridor or an unvented area.
2. The method of claim 1, wherein determining the support state, the occupancy state, and the to-be-verified state of the local spatial unit comprises: Determining depth information, residual motion information and shielding boundary information of the local space unit according to the multi-view time sequence image and the pose incremental data; determining surface continuity of the local space unit according to the depth information; determining an occupation relationship and an occlusion relationship of the local space unit according to the residual motion information and the occlusion boundary information; And determining a supporting state, an occupying state or a state to be verified of the local space unit according to the surface continuity, the occupying relation and the shielding relation.
3. The method of claim 2, wherein the determining the support state, the occupancy state, or the to-be-verified state of the local spatial unit based on the surface continuity, the occupancy relationship, and the occlusion relationship comprises: determining that the local space unit is in a supporting state under the condition that the gradient, the height difference and the surface roughness of the local space unit meet preset passing conditions; Determining that the local space unit is in an occupied state under the condition that the residual motion intensity and contour continuity of the local space unit meet preset occupation conditions; And determining the local space unit as a state to be verified under the condition that the local space unit is positioned behind the depth discontinuous boundary determined according to the depth information and the sight is blocked.
4. A method according to claim 3, wherein said constructing a three-state spatial profile comprises: marking the supporting state, the occupying state and the state to be verified according to the position relation among the local space units; Merging the local space units which have the same state and are adjacent to each other into a state area; and forming the three-state space distribution diagram according to the state area.
5. The method of claim 1, wherein the generating a candidate pass corridor comprises: Sweeping the three-state space distribution map along a candidate direction based on the robot body envelope to form a plurality of passing units; Determining a passing unit corresponding to the current position of the robot body envelope as an initial passing unit; Determining a support continuity authentication result according to a support state corresponding to the passing unit, determining a passing clearance authentication result according to a distance relation between the robot body envelope passing through the passing unit and a local space unit in an occupied state, determining a connectivity authentication result according to an adjacent relation between the passing unit and the initial passing unit, and determining a time window authentication result according to a dynamic obstacle occupation time window corresponding to the passing unit; And generating the candidate traffic corridor according to the support continuity authentication result, the traffic clearance authentication result, the connectivity authentication result and the time window authentication result.
6. The method of claim 5, wherein the dynamic obstacle occupation time window is used to characterize a time interval in which the dynamic obstacle occupies a corresponding traffic unit; The determining the time window authentication result includes: Determining an estimated arrival time period for reaching the passing unit according to the motion constraint corresponding to the robot body envelope; comparing the estimated time of arrival with the dynamic obstacle occupation time window; Determining that the passing unit meets an instant passing condition under the condition that the estimated arrival time period is not overlapped with the dynamic obstacle occupation time window; And determining that the passing unit meets a time window passing condition under the condition that the expected arrival time period overlaps with the dynamic obstacle occupation time window and a conflict-free passing time period exists.
7. The method of claim 6, wherein the generating the candidate pass corridor further comprises: Merging the continuous adjacent passing units meeting the supporting continuity authentication result, the passing clearance authentication result and the connectivity authentication result and meeting the instant passing condition or the time window passing condition to obtain a candidate communication area; and merging the traffic units which are adjacent to the candidate communication area and correspond to the local space units in the state to be verified into the candidate communication area to obtain the candidate traffic corridor.
8. The method of claim 1, wherein the active supplemental observation action comprises: Determining the position of a passing unit corresponding to the local space unit in the state to be verified in the candidate passing corridor; selecting a target active supplementary observation action from a preset action set according to the position; executing the target active supplementary observation action to acquire the supplementary observation image; the set of preset actions includes at least one of a steering in place action, a lateral translation action, and a low speed forward action.
9. The method of claim 8, wherein the projection consistency check comprises: determining an obstacle profile, a depth discontinuity boundary and a dynamic obstacle profile according to the supplemental observation image; Projecting the robot body envelope to the supplementary observation image along the candidate passage corridor to obtain a body projection profile; And determining a projection consistency check result of the candidate passing corridor according to an overlapping relation or an intersecting relation between the body projection profile and the barrier profile, the depth discontinuous boundary and the dynamic barrier profile.
10. The method of claim 9, wherein the updating the three-state spatial profile and outputting an instantaneous traversable corridor, a time window traversable corridor, or an unvented region comprises: under the condition that the projection consistency check result represents that the candidate passage corridor meets consistency conditions, updating the to-be-verified state local space unit corresponding to the candidate passage corridor into a supporting state local space unit; Updating the local space unit to be verified corresponding to the candidate passage corridor into an occupied state local space unit or keeping the local space unit to be verified under the condition that the projection consistency check result indicates that the candidate passage corridor does not meet the consistency condition; and outputting the instant passable corridor, the time window passable corridor or the non-passable area according to the updated three-state space distribution diagram and the passing unit authentication result.

Description

Dynamic obstacle scene vision passable area identification method Technical Field The invention relates to the field of robot vision perception, in particular to a method for identifying a vision passable area of a dynamic obstacle scene. Background As mobile robot related industries continue to develop, equipment operational safety, traffic efficiency, and path planning stability have become fundamental capabilities. The existing visual navigation technology depends on static area judgment or single observation result, and when a shielding, moving target and local unobserved area are encountered, path interruption, false release or frequent re-planning are easy to occur, so that the continuous operation capability, the cooperative efficiency and the safety redundancy of the system are affected. Disclosure of Invention The invention provides a method for identifying a visual passable area of a dynamic obstacle scene, which is used for at least solving the problems that the conventional visual passable judgment is difficult to simultaneously treat a shielding area and a dynamic obstacle, so that the passable area is misjudged and a path decision is unstable. The invention provides a method for identifying a visual passable area of a dynamic obstacle scene, which comprises the following steps: According to the multi-view time sequence image and the pose incremental data, determining the supporting state, the occupying state and the state to be verified of the local space unit, and constructing a three-state space distribution diagram; Based on the robot body envelope and the dynamic obstacle occupation time window, carrying out support continuity, traffic clearance, connectivity and time window authentication on traffic units in the three-state space distribution diagram, and generating candidate traffic corridor; Under the condition that a local space unit in a state to be verified exists in a traffic unit corresponding to the candidate traffic corridor, executing an active supplementary observation action, acquiring a supplementary observation image, and carrying out projection consistency verification on the candidate traffic corridor based on the supplementary observation image; And updating the three-state space distribution diagram according to the passing unit authentication result and the projection consistency check result, and outputting an instant passable corridor, a time window passable corridor or an unvented area. In one possible implementation, the method for determining the support state, the occupation state and the state to be verified of the local space unit comprises the steps of determining depth information, residual motion information and shielding boundary information of the local space unit according to multi-view time sequence images and pose increment data, determining surface continuity of the local space unit according to the depth information, determining occupation relation and shielding relation of the local space unit according to the residual motion information and the shielding boundary information, and determining the support state, the occupation state or the state to be verified of the local space unit according to the surface continuity, the occupation relation and the shielding relation. In one possible implementation, the method for determining the supporting state, the occupying state or the state to be verified of the local space unit according to the surface continuity, the occupying relation and the shielding relation comprises the steps of determining the local space unit as the supporting state under the condition that the gradient, the height difference and the surface roughness of the local space unit meet preset passing conditions, determining the local space unit as the occupying state under the condition that the residual motion strength and the contour continuity of the local space unit meet preset occupying conditions, and determining the local space unit as the state to be verified under the condition that the local space unit is located behind a depth discontinuous boundary determined according to the depth information and the sight is shielded. In one possible implementation, constructing the three-state spatial distribution map includes marking a support state, an occupied state, and a state to be verified according to a positional relationship between local spatial units, merging the local spatial units having the same state and adjacent to each other into a state region, and forming the three-state spatial distribution map according to the state region. In one possible implementation, generating a candidate passage corridor comprises sweeping a three-state space distribution diagram along a candidate direction based on a robot body envelope to form a plurality of passage units, determining the passage unit corresponding to the current position of the robot body envelope as an initial passage unit, determining a support continuity authentication resu