The Research And Implementation Of Multi-Depth Cameras Based Pedestrian Counting And Collaboration Analysis

As one of the main function of intelligent video surveillance, pedestrian counting has important research value, and can be widely used in crowded scenes such as shopping mall, side walk, subway exit, and so on. The traditional pedestrian counting methods mainly used a single normal RGB camera to realize accurate individual counting, but the results were always influenced by factors such as background, illumination, wind, shadow, and so on. Moreover, non-rigid feature of human body and overlapping problem also was not resolved yet. Our method is based on a new depth camera, which can easily resolve these problems. We also expands from a single camera to multiple ones. Therefore, the monitoring area is no longer limited, and the collaboration analysis between the multiple regions can obtain pedestrian movement rules in the whole scene. This analysis can make it possible to model crowded pedestrian movement.The paper proposes a pedestrian counting method with the combination of depth information based on depth camera. On the basis of single camera method, we further research on the collaboration analysis of multi-depth cameras. For linear distribution of multiple cameras, we can normalize them into one common coordinate system through coordinate conversion, then the counting procedure becomes the same with the method based on a single camera. On the contrary, for non-linear distribution of multiple cameras, we detect, track and count each region, respectively. Then we can model the whole pedestrian scene through collaboration analysis.Our experiments in different scenes demonstrate that our proposed pedestrian counting and collaboration analysis method based on depth cameras can obtain exciting results. Meanwhile, it can be used in real time. And in the future, it may bring high application value and great development space.