Three-dimensional target detection method and device
By fusing features from point cloud data and image data at the target level and employing dynamic weights based on feature confidence, this method addresses the issues of high computational resource overhead and low detection accuracy in existing target detection methods. It achieves real-time and lightweight target detection, making it suitable for complex scenarios.
Patent Information
- Application Number
- CN202511009784.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-22
- Publication Date
- 2025-10-17
AI Technical Summary
Existing target detection methods have high computational resource overhead, slow inference speed, inaccurate spatial alignment, lack of fine modeling capabilities for key targets, and poor detection accuracy and stability. They are difficult to meet the requirements of real-time and lightweight application scenarios, and are difficult to adapt to actual scenarios with complex occlusions, multi-scale targets or background interference.
By acquiring spatiotemporally synchronized point cloud data and image data within the area to be detected, representative point cloud target boxes and representative image target boxes are selected respectively. The matching point cloud features and image features are fused using dynamic weights based on feature confidence to obtain fused features, and finally, the 3D target detection result is obtained.
It reduces computational resource consumption, improves inference speed, enhances spatial alignment accuracy, enables detailed modeling of key targets, adapts to scenarios with occlusion, multi-scale or complex background interference, improves detection accuracy and stability, and meets the needs of real-time and lightweight application scenarios.