一种无需数据标注的开放点云目标检测方法及系统
By employing an open point cloud target detection method that does not require data annotation, this method utilizes point cloud clustering and contrast loss to increase intra-class similarity and decrease inter-class similarity. Combined with a combined prediction strategy, it achieves unlimited category detection of railway obstacles, solving the problems of difficult data acquisition and limited categories in existing technologies, and improving the accuracy and adaptability of detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- STATE GRID JIANGSU ELECTRIC POWER CO LTD NANJING POWER SUPPLY COMPANY
- Filing Date
- 2022-11-25
- Publication Date
- 2026-07-17
AI Technical Summary
Existing target detection methods rely on data labeling for railway obstacle detection, which makes data collection difficult and fails to identify obstacles in the real world that are not labeled with categories, resulting in the risk of missed detections. In particular, it is difficult to achieve effective detection in low-probability anomalies on railway tracks.
An open point cloud target detection method without data annotation is adopted. By establishing an open scene point cloud foreground target detection network with dual detection heads, point cloud clustering and contrast loss are used to increase intra-class similarity and reduce inter-class similarity. Combined prediction strategies are used to predict unknown obstacles.
It enables the detection of an unlimited number of obstacles in a railway track scenario, improving the accuracy and adaptability of obstacle recognition and solving the challenges of insufficient data and limited categories.
Smart Images
Figure CN115984594B_ABST