Person re-identification based on minimum feature using calibrated camera

Although several approaches of person re-identification can be found in the paper, tracking people restricted in an open area is still an active research. In this paper, we propose a method assuming each pedestrian as a collection of multiple elements of the database. From the lowest point for the sector, layer by layer up to the top, we collect the minimum feature of the object. The features include statistic feature based on Bhattacharyya distance, SURF feature and Color histogram. After the pedestrian identification, and then start tracking. All the patches can be tracked individually and the vectors are calculated with the world coordinate. Not only that, the algorithm also uses the calibrated parameters of multi-camera to directly compute the pedestrian scale, and at the same time limit the region of searching, rather than using all the complex features as before. The experiment is carried out using two sets of publicly available database (PETS and CHUK). According to the experiment results, our method has a strong robustness, even for the larger changes in the scale of person can also be identified. Algorithm can also deal with many occasions, the accuracy is also high, compared to many of the existing state-of-art algorithm.