Joint Customer Pose and Orientation Estimation Using Deep Neural Network from Surveillance Camera
Jingwen Liu, Yanlei Gu, Shunsuke Kamijo · 2016
The analysis of customer pose and orientation from surveillance camera is one of the most important topics for marketing. In order to understand the customer interest level to the merchandise, retailers are most concerned the customers near the merchandise shelf. In those areas, the customer pose and orientation are highly dependent on each other. Therefore, we propose a joint customer pose and orientation estimation system from surveillance video by using deep Neural Network. This system composes the dependency between pose and orientation. In addition, considering the customer pose and orientation changes gradually over time, we apply a prediction structure to improve the system performance. In this structure, the estimated pose and orientation of one frame are both used to enhance the estimation of next frame. We also propose a two-step learning algorithm to train this network. At last, we conduct a series of comparison experiments. The experimental results also prove the effectiveness of our proposed system.