Traffic Congestion Net (TCNet): An Accurate Traffic Congestion Level Estimation Method Based on Traffic Surveillance Video Feature Extraction

Traffic congestion is a common traffic anomaly in many large-scale cities. The research on video-based traffic congestion evaluation methods is the main trend of traffic congestion detection, but its discrimination on the level of congestion remains to be studied. In this paper, we introduce a traffic congestion estimation method based on traffic surveillance video feature extradition named TCNet (traffic congestion net). Taking the density and speed of traffic flow as the congestion estimation standard, the vehicle detection and speed estimation as the technical core, TCNet can provide a more accurate description of the congestion level. TCNet uses improved YOLOv3 module to detect the image to get the number of vehicles on the road, using TBBFA (traffic bounding box filtering algorithm) to remove the redundant and error bounding boxes, thereby getting the accurate traffic flow density. Finally, we use the TCMA (timing-based center matching algorithm) to calculate the driving speed measured by pix/sec of each detected vehicles. With the above-calculated parameters, we can finally calculate the level of congestion. For practical application, TCNet's detection time is optimized to achieve the effect of real-time surveillance detection.