Automatic Moving Object Extraction Through a Real-World Variable-Bandwidth Network for Traffic Monitoring Systems

Automated motion detection has become an increasingly important subject in traffic surveillance systems. Video communication in traffic surveillance systems may experience network congestion or unstable bandwidth over real-world networks with limited bandwidth, which is harmful in regard to motion detection in video streams of variable bit rate. In this paper, we propose a unique Fisher's linear discriminant-based radial basis function network motion detection approach for accurate and complete detection of moving objects in video streams of both high and low bit rates. The proposed approach is accomplished through a combination of two stages: adaptive pattern generation (APG) and moving object extraction (MOE). For the APG stage, the variable-bit-rate video stream properties are accommodated by the proposed approach, which subsequently distinguishes the moving objects within the regions belonging to the moving object class by using two devised procedures during the MOE stage. Qualitative and quantitative detection accuracy evaluations show that the proposed approach exhibits superior efficacy when compared to previous methods. For example, accuracy rates produced by F-1 and Similarity metrics for the proposed approach were, respectively, up to 92.23% and 88.24% higher than those produced for other previous methods.