Research On Video Based People Counting Method In Indoor Spaces

For the purpose of counting people in different indoor spaces (such as corridors, rooms), the technology of people counting based on video image analysis is studied. It makes statistics of the number of people that are in and out of the camera view on a certain period of time or in the designated area at certain time, to provide data for the energy saving control, business management, and so on.By analyzing the main technical route and classic algorithm of people counting method and people crowd density estimation, a people detection algorithm based on whole body HOG feature is designed. To overcome the disadvantage of slow speed of original HOG people detecting, the indoor environment and targets size are used as constrains to improve people detecting speed of the proposed algorithm. An analysis method of people position attribute is proposed based on a combination of a tripwire and linear discriminant function. This method can estimate whether people cross the tripwire and its moving direction, to get the number of people that come and go. Considering the occlusion from peoples and environmental facilities, a people detection algorithm based on head-shoulder HOG model is researched. The key technique of the transplanting of the proposed people counting algorithm to the Android platform is researched, then the technology framework of distributed people counting sensor based on Android is designed.