Affiliations: [a] Xian Key Laboratory of IOT Engineering, Xian University, Xian, PR China
| [b] Ressearch Center of Hubei Logistics Development, Hubei University of Economics, Wuhan, PR China
| [c] School of management, Xian Jiaotong University, Xian, PR China
| [d] School of engineering, Tibet University, Lasa, PR China
Abstract: This paper presents a Feedback Visual Background extractor (FViBe) method for salient foreground detection in urban traffic scenes to efficiently resolve deficiencies that the background subtraction model is easily contaminated by temporarily stopped or slow motions objects. The background template is constructed based on the history of recently observed pixel samples and each pixel is assigned counters to describe the corresponding traffic state and stability. The threshold for salient foreground decision is set adaptively according to stability of scenes, and model update depends on the feedback current traffic state and the stability. The overall results obtained with the real-world urban traffic videos are presented to demonstrate that the FViBe achieves better performance of both visual comparison and quantitative evaluation than other state-of-the-art methods, particularly in the slow motions or temporarily stopped objects traffic scenario. Moreover, the experimental results show that the FViBe is suitable for real-time implementation in salient foreground detection of the urban traffic scenario.
