Pedestrian re-identification method and system under occlusion dressing scene
By using an image-text comparison model and a set of structured text prompt templates to locate occluded parts, generating a visibility mask for feature completion, and employing a dual-branch collaborative learning structure to suppress clothing bias, the system addresses the problem of insufficient robustness in pedestrian re-identification under occlusion and clothing-changing scenarios, achieving high-precision pedestrian recognition.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUBEI UNIV
- Filing Date
- 2026-03-10
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies lack a collaborative mechanism in pedestrian re-identification in scenarios involving occlusion and clothing changes, resulting in insufficient robustness and difficulty in accurately identifying pedestrians in complex real-world scenarios.
An image-text comparison model and a two-level structured text prompt template set are used for occlusion localization. A visibility mask is generated for feature completion. A two-branch collaborative learning structure is used to suppress clothing-related bias and output pedestrian re-identification results in occlusion changing scenes.
It significantly improves the robustness and reliability of pedestrian re-identification methods in complex real-world scenarios, enabling accurate identification of pedestrians even under the dual interference of occlusion and clothing changes.
Smart Images
Figure CN122416518A_ABST