Weight determination in multi-feature fusion for pedestrian re-identification

Pedestrian re-identification (Re-ID) via cross-camera is a difficult problem in the field of pedestrian discovery and tracking. Traditional solutions rely heavily on the external characteristics of a pedestrian's appearance. Maximally Stable Color Regions (MSCR), RGB (Red, Green, and Blue), HSV (Hue, Saturation and Value), and Histogram of Oriented Gradient (HOG) are usually used features. However, a single feature often cannot get a good matching result Multi-feature fusion has become a preferred method. This approach often requires determining the proportion of various features in the fusion process. In this paper, a method is proposed and used to confirm the proportion of various features at the optimal matching rate. The experiments are operated on a well-known dataset. Finally, the best weight combination of MSCR, RGB, wHSV, and HOG was determined by the proposed method.