A Fast Vehicle Detection Method for Road Monitoring

Abstract

To address the issue that current traffic monitoring system needs to automatically detect the objects on the road, a fast vehicle detection method was proposed. Applied the YOLO framework, this paper considered the vehicle detection as a regression problem of vehicle location prediction and classification prediction. Under the precondition of guaranteeing accuracy, the structure of convolution neural network (CNN) is optimized to speed up the detection. The vehicle model CarNet was trained on a dataset containing a large number of road vehicle sample images. The experimental results show that with this method vehicle in video can be detected quickly even on a computer without GPU. The method also shows good robustness and high precision.