Approach to Auto-Recognition of Human Trajectory in Squares using Machine Learning-Based Methods: An application of the Yolo-v3 and the DeepSORT algorithm

Abstract

The square plays an essential role in contemporary urban space. Researchers had explored many methods to record the distribution of people in it. However, few of them study this issue using fine data. This study proposes a method recognized of the human trajectory using a machine learning-based computer vision algorithm, which can be divided into the following three steps. (1) the acquisition of video and the method of obtaining human trajectory. (2) cleaning of the raw human trajectory data. (3) to visualize the trajectory data. Based on the existing methodology, we take three example squares within the Tianjin University campus to illustrate it. We use trajectory map, people distribution heat map, and people walking speed heat map as visualization methods. The following two conclusions are drawn from the three examples. First, it is found that the human trajectory data derived from this method is more accurate when the UAV is flying at a lower altitude. Second, this study demonstrates that a passive Real Time Locating Systems (RTLS), based on a deep learning computer vision algorithm, can effectively obtain human trajectory data in a square. Third, this paper proves that the visualization method we used is effective.