Real-Time Traffic Spatial Occupancy Calculation with Modified YOLOv3 in Complex Environments

The number of vehicles on the road continues to rise due to population growth, economic development, urbanization, technological advancements, and improved infrastructure. The result is that traffic congestion, delays, and emissions are on the rise in cities. A deep learning-based technology called Traffic signal control system (TSCS) is being used to deal with these issues at intersections. Through the TSCS, accidents, frustrations, and impatience can be reduced by efficiently regulating traffic flow. A TSCS’s effectiveness in terms of green signal time is determined by both vehicle counts and the amount of space vehicles occupy. Traffic counting and spatial occupancy are calculated accurately and quickly with accurate and fast vehicle detection and tracking. This study presents a framework for calculating spatial occupancy as well as a comparative analysis of vehicle counting in each cycle. Two trackers are used in this framework: traditional Euclidean distance and deep sort, which are designed to count different types of vehicles in complex environments like QMUL junctions, along with modified Yolo v3. Our experiment showed that deep sort performs far better than the traditional Euclidian distance tracker. The MAE of the modified Yolo v3 with deep sort is 3.0, whereas the MAE of the traditional Euclidian distance tracker is 11.8. The results of this analysis study may result in an improvement in traffic management and congestion control in the future.