A Multi-object Detection Method Based on Connected Vehicles

Nowadays, with internet of vehicle developing, more and more research institute begin to research intelligent transportation systems. Among these systems, vehicle-road collaborative system is a prominent one. Its main function is to percept traffic situation. And using image recognition technology is one of the methods. The advantages of this method are low cost, high data-correcting rate, and small interference to traffic flow. Traditional image recognition algorithms always have problems with high-processing time and low accuracy, such as HOG (Histogram of Oriented Gradient) and DPM (Deformable Part Models). They are not suitable to monitor real-time traffic videos of numerous image frames. In this paper, a new model is proposed based on the original YOLOv3 algorithm. Compared with original YOLOv3 algorithm, the proposed model in this paper can not only realize multi-object detection, but also consume less time and recognize more accurately. Moreover, this paper constructs a special identification data set of common traffic participants (pedestrians, cars, buses, etc.) to monitor traffic flow, based on several data sets, such as INRIA, MIT and KITTI. Moreover, this paper collects images of commonly-seen objects at traffic intersections to test the performance of the proposed model in this paper.