Video-CRM: Understanding Customer Behaviors in Stores

This paper describes a real-time computer vision system to detect and track people in stores to understand retail customer behavior while shopping. We propose an approach to detect and track people and their body posture without using an explicit 3D human model using overhead narrow-baseline stereo cameras. The proposed method is based on a 3D silhouette of people that is constructed from a 2D silhouette. The 2D silhouette is detected by color and disparity background subtraction. Once the 3D silhouette is generated, people are identified with iterative segmentation of a 3D silhouette based on the topological structure of human body. Once people are detected, their appearance model based on color and shape are generated and tracked over multiple camera using their 2D trajectory continuity and their appearance models. A shape histogram, the distribution of relative positions of points on a 3D silhouette, is introduced to estimate the posture and body parts to understand the people and object interactions in stores, such as "customer picking a object from a shelf". The initial pilot studies show the real-time performance and accuracy of the
system to understand customer behavior.