A passenger flow statistic algorithm based on machine learning

For the bad accuracy and high costs of the transit passenger traffic statistics, we propose a machine vision-based algorithm. As the body parts of the human often occur overlap in congested conditions, the algorithm in this study uses the overhead image of the passengers as the statistic targets. The Histogram of Oriented Gradient (HOG) combined with the Support Vector Machine (SVM) is adopted in this study. Compared with the traditional methods, the accuracy and the speed of the detection stage have been greatly improved. The algorithm can be divided into three steps: the target detection, target tracking and statistics. Based on the machine learning methods, it first extracts the head HOG feature, and then uses the SVM algorithm to train feature sets to get a classifier based on passenger heads. We choose the Camshift as the tracking algorithm. In order to improve the correct trajectory, it must be fuse the target detection and the target tracking. Through the rules of the passengers to alight, we can get the number of passengers automatically.