A method and system for multi-view target detection at intersections based on corner pooling

Abstract

The invention discloses a multi-view target detection method and system at an intersection based on corner point pooling. The method includes: preprocessing the images of the multi-view camera at the intersection collected in real time; Input the pre-established and trained multi-view target detection model at the intersection, and output the target prediction result; wherein, the multi-view target detection model is used to extract the features of the preprocessed multi-view camera image, and perform feature projection on the extracted features , feature fusion and corner pooling, predict the target position through the ground plane rectangular feature map after corner pooling processing, and at the same time perform single-view detection and result projection on the extracted features, and perform target position through the single-view target position map Correction, output the target prediction result.

Description

technical field [0001] The invention belongs to the field of target detection, and in particular relates to a method and system for detecting a multi-viewpoint target at an intersection based on corner point pooling. Background technique [0002] With the rapid development of unmanned driving and smart cities, single-sensor vehicle detection technology has been relatively mature. However, at intersections with complex traffic conditions, dense detection caused by vehicle congestion is difficult, and occlusion problems caused by bulky vehicles, Factors such as the uncertainty of a single sensor severely restrict the accuracy of vehicle detection, and there are also safety hazards in complex intersections. With the introduction of multi-view detection methods, the detection performance of intersection vehicles in crowded or occluded scenes has been significantly improved, which greatly promotes the safety of unmanned driving. However, vehicle detection methods based on multi-...

AI Technical Summary

Problems solved by technology

However, feature fusion can only reduce the amount of calculation, and does not essentially improve the "ghosting" phenomenon, but detects two ghosts as a larger target, which will also interfere with the final decision.

Images