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.

CN120808334APending Publication Date: 2025-10-17BEIJING CENTURY DONGFANG COMMUNICATION EQUIPMENT CO LTD
0 Cites 1 Cited by

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

Technical Problem

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.

Method used

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.

Benefits of technology

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.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to the technical field of target detection, and provides a three-dimensional target detection method and device. The method comprises the following steps: acquiring time-space synchronized point cloud data and image data in a to-be-detected area; respectively screening representative target frames from the point cloud data and the image data; matching the point cloud features and the image features in the representative target frame to obtain matched point cloud features and matched image features; fusing the matched point cloud features and the matched image features by adopting a dynamic weight based on feature confidence to obtain fused features; and obtaining a three-dimensional target detection result based on the fusion features. The target-level fusion is matched with the fusion weight adjustment based on the feature confidence, so that the computing resource overhead can be reduced, the reasoning speed can be increased, the space alignment accuracy can be improved, the key target can be modeled finely, and the detection precision and stability can be improved, and therefore, the application scene requirements of real-time performance and light weight can be met; and the method is suitable for shielding, multi-scale or background interference complex scenes.
Need to check novelty before this filing date? Find Prior Art