An Effective Vehicle Counting Approach Based on CNN
Linjun Yao · 2019
Vehicle counting is an essential part of the intelligent transportation system (ITS). Compared with traditional vehicle detection methods, convolutional neural network (CNN) could reach in real-time speed with high accuracy in vehicle detection. This paper presents an efficient and robust method of vehicle counting by combining CNN and the virtual coils. Experimental results indicate that our approach could remove duplicate and missing detection effectively and count the vehicle at high speed with around 98% relative accuracy and 93% absolute accuracy on average while it could also work well in the traffic jam situation with over 95% relative and 84% absolute accuracy. Besides, through identifying the traffic light color, the proposed method could also be applied in automatically capturing red light violations.