Vehicle Speed Monitoring using Convolutional Neural Networks
Keywords:
CNN, traffic analysis, urban traffic, overspeed detection, computational vision.Abstract
The usage of computer vision techniques applied to security cameras video has motivated developing robust traffic monitoring applications. Traditional methods use frames subtraction algorithms to detect moving vehicles. The main disadvantage of those methods, for urban roads surveillance, comprises its poor ability to handle lighting changes or pedestrians on the video frame. Unlike existing methods, this paper proposes using Convolutional Neural Networks (CNN) for both background segmentation and the recognition of regions of interest applied to monitor vehicle speed. It applied the CNN based method to different datasets to evaluate the developed application. The results reveal 87% hit rate. Although the overall accuracy is near to background subtraction based methods, the major advantage of this approach is being environmentally adaptable which may significantly impact the design and analysis of new traffic monitoring systems.