Study on Method of Multi-feature Reduction Based on Rough Set in Passenger Counting

In view of the instability and inaccuracy of the existed automatic passenger counting method based on the video sequence, in order to improve the instantaneity and effectiveness of the statistic method, this paper presented a new way based on multi-feature. It used single camera vertical shooting way to get the video, and extracted the moving targets in foreground by Gaussian mixture background model. A series of representative features to the foreground were extracted, such as aspect ratio, rectangularity, circularity, corner points of Shi-Tomasi, Hu invariant moments Fourier descriptors and LBP descriptors, after training these characteristics respectively, retained the high precision features and combined them all, in order to counting the number of people, this paper trained these features based on the neural network. In the experimental process, it evaluated the proposed method with several criterions, Experimental results show that the proposed algorithm has a high statistical precision and good efficiency.