CNN Traffic Detection Method Based on Adaptive Context Information

Abstract

The invention discloses a CNN traffic detection method based on adaptive context information, including a training stage and a detection stage, in a specific traffic scene, training an adaptive context feature selection model; on the basis of obtaining the adaptive context feature selection model First, train the CNN traffic detection system based on adaptive context information; in the forward stage, through post-processing, accurately frame the traffic target (during detection, perform joint prediction of context and target, and accurately frame the traffic target through post-processing). The present invention proposes a CNN traffic detection system based on adaptive context information, which mainly includes a CNN-based adaptive context selection model and a traffic detection system fused with the model, further improving the accuracy of vehicle and pedestrian detection.

Description

technical field [0001] The invention relates to a CNN traffic detection technology based on adaptive context information that can be applied in real time. [0002] Background technique: [0003] In order to solve these increasingly serious traffic problems, intelligent transportation systems (Intelligent Transportation Systems, referred to as ITS) came into being, in which vehicle and pedestrian recognition is an important component of intelligent transportation systems, and some existing related technologies about vehicles and pedestrians have also been obtained. widely used. [0004] The current traffic detection system mainly realizes the recognition and detection of different targets by describing the appearance information of the targets (pedestrians and vehicles). At present, this type of system mainly uses artificially designed features (such as HOG, LBP, SIFT, etc.) or deep features obtained directly from the image itself through deep learning to describe the appeara...

AI Technical Summary

Problems solved by technology

However, in the actual detection of daily traffic, most of them are unconstrained open environments, complex and changeable, and there are interferences such as illumination changes, perspective changes, and target occlusions. If we only start from the appearance information of the target itself, when images or videos When the information provided by the traffic target itself is too little, the target category cannot be accurately judged only based on the target itself

Moreover, there are certain differences in different traffic scenes, and a pervasive traffic object detection system that ignores the differences in different traffic scenes will reduce the accuracy of traffic object detection

Images