Real-Time Vision-Based Pedestrian Detection By Contour Matching

Dhaka University Journal of Applied Science & Engineering

Title:	Real-Time Vision-Based Pedestrian Detection By Contour Matching
Authors:	Mohsin Khan Department of Computer Science and Engineering, University of Dhaka, Dhaka-1000, Bangladesh Emon Kumar Dey Department of Computer Science and Engineering, University of Dhaka, Dhaka-1000, Bangladesh Md, Hasanuzzaman Department of Computer Science and Engineering, University of Dhaka, Dhaka-1000, Bangladesh
DOI:	PDF
Keywords:	Motion Areas Segmentation, Scan Conversion Algorithm, Contour, Pedestrian Detection
Abstract:	Real time Pedestrian detection is a challenging and important task for computer vision-based researchers. This paper presents a novel method for real-time pedestrian detection. In this method motion areas are segmented from real-time image frames captured by a CCD camera. From the segmented and filtered motion area contour is detected using proposed scan conversion algorithm. The normalized contour is then compared with predefined pedestrian shape contour templates and if satisfactory match is found then the pedestrian is localized and bounded using a rectangular box. The system is tested using 4960 image frames and achieved average precision and accuracy rates are 95.72% and 93.05% respectively.
References:	M. Enzweiler, D.M. Gavriia, "Monocular pedestrian I8l Detection: Suney and Experimenrs', IEEE Transaclion on Pan€m Analysrs and Machine ln(elligence (PAMl). Vol. 31(12), pp. 2179 -2195, 2001