Scene recognition in traffic surveillance system using Neural Network and probabilistic model
Duong Nguyen‐Ngoc Tran, Long Hoang Pham, Ha Manh Tran, Synh Viet‐Uyen Ha · 2017
In the traffic surveillance system (TSS), there are many factors affect the qualities of the result. Through practical application, it is difficult to determine which scene changing during the day period, from the daylight to nighttime, the conversion of the sunny and overcast, wet and dry scene. However, there have been no controlled studies which illustrate the method to distinguish environment scene, which is the one of six main challenges in TSS. Therefore, this paper presents the method to detect and recognize the change of scene during all-day surveillance; Thus, TSS adopt the recognition to determine the appropriate method for each scene, for increasing performance. Our recognition model is based on the combination of the CIE-Lab color space and the histogram of the region-of-interest (ROI) in each frame, which used for extracting the feature for the Feed Forward Neural Network to perform the detection. In the experiment section, our results show that the benefits of our proposed method in the real-world traffic surveillance system.