Human head detection using active basis model from surveillance videos

• Main Authors: Kuang Chiang, 江鑛
• Other Authors: 韓欽銓
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/42420213293026538394

碩士 === 國立聯合大學 === 資訊工程學系碩士班 === 102 === Analysis of people flow can be put into use in many ways, for instance, application of business such as counting people flow of gateway, according to the number of consumers to make further considerations of business policy. The effect of special promotions can be evaluated by comparing the number of consumers during promotional period and normal period. Further, after compared the proportion of consumers that really consumed would also be found, this information could be a datum to draw up selling methods which are more attractive. To classify the related works of people counting that by the mounted setting of camera, there would have two classes: mounted vertically and mounted in bevel, if camera was mounted in bevel, when people was not sheltered from other people and objects, the whole human body would be captured. In common situation, almost all cameras were mounted in bevel, so this paper focus on the situation which the camera is mounted bevel. This paper realized people counting by Active Basis Model, to detect crowd which is overlapping in the crowded event is a great challenge as people usually sheltered from each other, perhaps a part or whole, however, pedestrians would keep distance from each other, so the head in the overlapping crowd would not be sheltered easily. Head detection need to overcome the problem that the direction pedestrians faced is unstable. Stature, clothing and hairstyle increased the difference of each Individual, moreover, such as head touching and waver would affect the performance of detection. Active Basic Model is an algorithm that can be in accordance with shape of objects to make model change, and adapt variations actively. The experiment result proved the well performance of crowded pedestrian detection in the approach which is proposed by this paper, and this result can be used in the people counting system.