This paper presents a vehicle behavior analysis method of the real-time video streaming based on MobilenetYOLOV4 and ERFNet. Firstly, an image dataset SYGData based on roadside camera is established, which contains a variety of vehicles and vehicle behavior to keep consistent with the actual traffic scene. Secondly, we present a new deep learning model Mobilenet-YOLOV4 which backbone is Mobilenet network to improve detection accuracy and speed. Compared with YOLOV4, its mAP increased from 77.79% to 84.53%, and inference speed increased from 4 frames per second to 25 frames per second. The results demonstrate its performance improvement in the accuracy and real-time over YOLOV4. Thirdly, we present a FFmpegOpenVINO framework that supports multi-model driving. It loads Mobilenet-YOLOV4 and ERFNet to detect vehicles and lane lines. Finally, the detection data are used to analyze vehicle behavior, including vehicle tracking, over-speed detection, lane change detection and traffic statistics. The results show that this method is feasible. It not only improves the detection accuracy and speed but also can analyze various vehicle behaviors on an intelligent roadside device.
Method of Vehicle Behavior Analysis for Real-Time Video Streaming Based on Mobilenet-YOLOV4 and ERFNET
11.11.2022
1176762 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
EUROPEAN RADIATION FACILITIES NETWORK (ERFNET)
TIBKAT | 2020
|Real-Time Vehicle Detection Based on YOLOv4 Neutral Network
Springer Verlag | 2022
|Enhanced ERFNet Decoder for Road Segmentation Model
TIBKAT | 2021
|