Pedestrian detection using a mixture mask model

Pedestrian detection is one of the fundamental tasks of an intelligent transportation system. Differences in illumination, posture and point of view make pedestrian detection confront with great challenges. In this paper, we focus on the main defect in the existing methods: the interference of the non-person area. Firstly, we use mapping vectors to map the original feature matrix to the different mask spaces, then using a part-based structure, we implicitly formulate the model into a multiple-instance problem, and finally use a MIL-SVM to solve the problem. Based on the model, we design a system which can find pedestrians from pictures. We give detailed description on the model and the system in this paper. The experimental results on public data sets show that our method decreases the miss rate greatly.