Dual-mode cooperative target detection suppression optimization method and equipment

Abstract

The invention aims to provide a dual-mode cooperative target detection suppression optimization method and device, and distance measurement can be introduced in combination with position information predicted by a model. The method comprises the following steps: firstly, eliminating similar redundant bounding boxes of the same target through an intra-class suppression mode by utilizing an intersection-to-parallel ratio IOU and a distance between bounding boxes, and then designing a distance-based inter-class suppression method for eliminating an error class bounding box predicted by a model for the same target. According to the method, the distance measurement is introduced, the advantages of an intra-class inhibition mode and an inter-class inhibition mode of the NMS algorithm are combined, a redundant bounding box predicted by the model is inhibited more reasonably, the problem of unreasonable inhibition of a traditional NMS algorithm is effectively solved, and the detection performance and the practical application capability of the model are enhanced.

Description

technical field [0001] The invention relates to the field of computers, in particular to a dual-mode cooperative target detection and suppression optimization method and device. Background technique [0002] Non-maximum suppression (NMS) algorithm is one of the most commonly used algorithms in target detection tasks, and its main purpose is to suppress redundant results of model detection, thereby outputting more correct results. In simple scenarios, the traditional NMS algorithm can achieve a good suppression effect, and can suppress the redundant detection results very effectively. [0003] The traditional NMS algorithm suppresses redundant detection boxes according to the Intersection over Union (IOU) between bounding boxes. like figure 1 As shown, assuming that the model now outputs two bounding boxes (bounding boxes, bbox) of A and B for a certain target, the confidence of A is P A , the area is S A ; the confidence level of B is P B , the area is S B ; the inters...

AI Technical Summary

Problems solved by technology

[0022] (1) When two objects belonging to the same category are in a close or rotating scene, NMS intra-class suppression is easy to perform false suppression operations on adjacent similar object bounding boxes;

[0023] (2) When the model has false detections for the same target, NMS intra-class suppression is difficult to eliminate redundant bounding boxes generated by false detections

[0025] In the inter-class suppression mode, the traditional NMS algorithm can eliminate the false detection boxes generated by the model for the same target, but when the target in the application scene is dense, close to each other, rotating, etc. The IOU is too high, it is easy to cause some false suppression problems, such as Figure 2f shown

[0026] To sum up, the traditional NMS algorithm is prone to some unreasonable suppression operations in complex scenes, and its suppression ability is limited, which is likely to cause problems such as missed detection and multi-frame false detection.

Images