目标检测后处理方法、装置、电子设备、车辆及存储介质
By performing confidence classification and scaling operations on the center point heatmap and target box size prediction results obtained in the target detection post-processing method, the problems of large data volume and long time consumption in traditional methods are solved, and more efficient and accurate target detection is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JILUO TECH (SHANGHAI) CO LTD
- Filing Date
- 2022-09-30
- Publication Date
- 2026-07-17
AI Technical Summary
Traditional target detection post-processing methods, such as NMS, involve processing large amounts of data, are time-consuming, and produce inaccurate results, leading to inaccurate final detection results.
By obtaining the original center point heatmap prediction results and target box size prediction results, the target points are classified according to their confidence level into sets with confidence levels greater than or less than a preset confidence threshold, and first and second target box sets are obtained respectively. Scaling and merging operations are performed before non-maximum suppression operations to avoid the problems of large data volume and long processing time.
This improves the efficiency and accuracy of target detection, reduces the amount of data required for nonmaximum suppression operations, and ensures the accuracy of the processing results.
Smart Images
Figure CN115578556B_ABST