A method of passengers detection inside the subway under complex circumstances

Passengers' comfort level is a quite important part for their consideration of out-going plan. A method of detecting subway's comfort level under complex circumstance is proposed in this paper. Our work consists of three main parts. The first is that we detect the number of passengers with the help of HOG (Histogram of Oriented Gradient) descriptor. This descriptor works well in pedestrian detection. The second achievement is that we train a kind of Adaboost classifier and improve it by cascade, which raise up the detection rate to a large extent. The third part is that we improve our work with camera calibration on the base of the first two part. We can quickly detect the number of passengers in the subway in a high accuracy according to our algorithm. Also, we are able to realize the real-time video detection.