Detecting Pedestrian Carried Objects in Surveillance Video

• Main Authors: Yueh-ju Tai, 戴悅如
• Other Authors: Mon-chau Shie
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/27338140265642044146

碩士 === 國立臺灣科技大學 === 電子工程系 === 101 === Automatic detection of carried objects by persons in video is an important step to be used for security monitoring, crime detection, and anti-terrorist surveillance. We propose a novel method using color segmentation to extract carried objects from pedestrians. We can detect not only position of objects but also their integral shapes. First of all, we can extract the pedestrian from each foreground image and further detecting pedestrian feature by taking advantage of skin detection and silhouettes symmetry analysis. After we obtain features of pedestrian, we present new watershed-based color segmentation to the extracted pedestrian target image. The goal of color image segmentation is to partition person image into several homogenous regions with similar properties. We use a new gradient image approach which combines edge detection with morphology and pedestrian features for the improvement of watershed algorithm. The gradient image achieves better color region segmentation in foreground image. We then extract region of carried objects from pedestrian using features which is detected before. Experimental results demonstrate that the proposed method is robust and accurate in detecting the shape, property and position of carried objects. Compared with other related thesis that we surveyed, our approach has better accuracy in the shape of carried object not an approximate position without shape information. Our method obtains a correct detection rate of 84% and also detect explicit region of carried object.