目标跟踪方法、装置、设备及存储介质
By generating adaptive template point clouds in sparse LiDAR scenes and combining them with a cross-frame aggregation module, the problem of target recognition difficulties in sparse scenes is solved, and fast and accurate target tracking is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- THE CHINESE UNIV OF HONG KONG (SHENZHEN) FUTURE NETWORK OF INTELLIGENCE INST
- Filing Date
- 2022-07-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing twin networks cannot effectively identify and track targets in sparse LiDAR scenarios, especially when the appearance of the target object changes rapidly. Existing methods cannot effectively enhance feature information in sparse scenarios, resulting in the inability to identify and track targets.
The system collects point cloud data sequences of the target scene using radar, determines the target's bounding box using a preset 3D target tracking algorithm, generates a first template point cloud, expands the template using an adaptive template generation algorithm, and combines historical information with a cross-frame aggregation module to generate a second template point cloud. The system extracts point cloud features for comparison and generates tracking results.
It enables rapid and accurate target localization, identification, and tracking in sparse scenes, eliminating noise interference and improving the accuracy and stability of target recognition.
Smart Images
Figure CN115272392B_ABST