Front-end pedestrian matching based on computational verb similarity

In the front-end of video monitoring, there is an important requirement that if a person occurring in one scene is the same guy as someone in the database. In this paper, we proposed an algorithm to measure that if two pictures we obtained contain the same person or not. We try different textures such as the vertical-horizontal histogram (VHH), Gray-Level Co-occurrence Matrices (GLCM) and Gabor wavelet (GW). Then we calculate the computational verb similarity with these features respectively for two different pictures. With a pedestrian database we set two indicators, Precision and Recall, to determine the final threshold similarity that two pictures contain the same person. And take the time consumption into consideration, we find GLCM and GW will spend much more time than the histogram. So we focus on the VHH and try to improve this feature. Considering the background in every pedestrian pictures, every picture is divided into different blocks. For every part we calculate the VHH respectively and get the average VHH of all blocks to calculate the verb similarity. Then the optimization is to find the best division parameter, the quantity of vertical blocks and horizontal blocks, which we can get the maximum sum of Precision and Recall.