Pre-processing framework with virtual mono-layer sequence of boxes for video based vehicle detection applications

Abstract

The increase in day to day vehicular traffic demands an efficient, automated and innovative approach for its management and regulation. One of the most practical and commonly used Intelligent Transportation Systems (ITS) based solution in recent times is video based traffic monitoring. The primary task of such systems is vehicle detection and the existing methods are not robust enough to handle diverse backgrounds and illumination changes in the traffic video. In order to address this issue, a Pre-processing Framework with Virtual Mono-Layered Sequence of Boxes (PF-VMSB) is proposed to make any vehicle detection process robust and efficient. The two composite modules of the proposed system include Effective Element Estimation (E3) module and Dynamic Multilevel Parallel Video Image Processing (D-MPVIP) module. The E3 module enhances the illumination of the traffic video only when necessary to produce an ideal environment for detection. Further, it classifies the type of background in video content (Static, Moderate or Dynamic) to provide an option to choose the appropriate image processing algorithm suitable for vehicle detection. In D-MPVIP module, the redundant video content are eliminated from processing to reduce the computational cost and processing time. Additionally, the spatial color information of the traffic video content is preserved and processed in parallel.The efficiency of the proposed framework of VMSB with E3 and D-MPVIP were analysed using benchmark datasets such as DETRAC, Urban Trackr, Ko-per and self recorded videos. The results shows an overall accuracy of detection rate using conventional image processing techniques such as Background Subtraction (BS) increases by 34.6% and the processing time is reduced by 37.5%. By incorporating the proposed framework into any vehicle detection process, the processing time and computational cost are improved without compromising the detection accuracy. The resultant of the detection process can be utilized by ITS application to enumerate traffic parameters such as vehicle volume count, congestion estimation, speed monitoring, travel time prediction, etc.