A Computationally Economic Novel Approach for Real-Time Moving Multi-vehicle Detection and Tracking toward Efficient Traffic Surveillance

详细信息    查看全文

• 作者：Raad Ahmed Hadi ; Loay Edwar George…
• 关键词：Adaptive search window ; Background subtraction ; Computational time cost ; Vehicle detection ; Tracking ; Traffic surveillance
• 刊名：Arabian Journal for Science and Engineering
• 出版年：2017
• 出版时间：February 2017
• 年：2017
• 卷：42
• 期：2
• 页码：817-831
• 全文大小：
• 刊物类别：Engineering
• 刊物主题：Engineering, general; Science, Humanities and Social Sciences, multidisciplinary;
• 出版者：Springer Berlin Heidelberg
• ISSN：2191-4281
• 卷排序：42

文摘

Efficient and precise real-time video-based vehicles supervision systems for traffic surveillance are ever demanding. Developing an accurate detection and tracking algorithm by incorporating the effects of vehicles shadow or visual obstacles/occlusion including road signs, trees, or other vehicles is a challenging task. We propose a new approach for efficient detection and tracking of moving vehicles in the presence of shadow and partial occlusion under complex road scenes by combining the modified background subtraction and innovative adaptive search window methods. The use of variable-size window (Bounding Box) reduces the searching space proportional to the allocated time of the moving vehicles. First, a background model is constructed and then subtracted from subsequent traffic frames to distinguish the foreground objects (moving vehicles regions) and stable background regions. Finally, the shadow detection and occlusion handling between the moving vehicles are implemented to leverage the surveillance performance. The performance of the proposed method is benchmarked using baseline highway category of changedetection.net standard dataset having 1231 frames. Performance evaluation is carried out in terms of time complexity analysis, detection accuracy, and processing time. The experimental results demonstrate a radical reduction in computational time in terms of searched space. A superior detection accuracy (more than 99%), and processing time \(\sim \)29.6 ms (per frames in milliseconds) together with the time complexity degree of O(\(n^{2}\)), is achieved. Excellent features of these results suggest that the proposed approach is prospective for computer vision application and high-way traffic surveillance.