Object Detection Based on Multi-Source Information Fusion in Different Traffic Scenes

In practical scenarios, object detection performance is largely affected by the current environment. For example, in traffic object detection scenes, object conglutination caused by the influence of light, occlusion and other negative factors brings difficulties to accurate multi-target detection. And the accuracy of object detection in traffic scenes directly impacts the safety of automatic driving. In this paper, a novel object detection method based on multi-source information fusion is proposed to deal with the low accuracy of object detection in different traffic scenes. The method first adopts a traditional two-stage detection network to extract visual features, then extracts the semantic features and relation features among objects, finally designs a multi-source information fusion module to integrate these features above. The fusion module actually enhances the learning ability of the network and improves detection accuracy. Moreover, the proposed network can achieve adaptability in various scenarios according to different geographic, environmental and weather characteristics. The simulation results show that the proposed approach has better performance in object detection than other traditional methods in different traffic scenes.