Attentive Billboards: Towards to Video based Customer Behavior
Ismail Haritaoğlu, Myron Flickner · Workshop on Applications of Computer Vision · 2002
We describe a real-time computer vision system and algorithmsthat extracts customer behavior information by detectingand tracking multiple people as they wait and watchadvertisements on a billboard or a new product promotionat a stand. Our system segments each frame into foregroundregions which contains multiple people. Foreground regionsare further segmented into individuals using a temporal segmentationof foreground and motion cues and global shapeconstraints on 2D Silhouettes. A 2D dynamic appearancetemplates is used to track people. The system can provideonline customer information, such as, number of peoplecurrently watching the billboard, their gender, and offlinecustomer data, such as, how long each people looked at thebillboard. Experimental results demonstrate robustness andreal-time performance of the algorithm.