Pedestrian re-identification is the key technology of intelligent video surveillance system. At present

most of the research on this technology is based on two-dimensional（ 2D） images. However

in practical application scenarios

the input video sequence is more suitable with the monitoring system application environment. According to optical flow energy value

video sequences which has better characteristic of distinguish is selected. HOG3D is used to extract the temporal dynamic features

combined with time feature

and fuse features of color and texture

combine with PFLMNN distance metric learning method

while reduce positive sample pair feature distance

at the same time

constrain the nearest neighbor negative sample pair feature distance. Experiments are carried out on public datasets i LIDS-VID and PRID-2011（ multi-shot）. Experimental results show that the proposed method achieves good matching precision on the two public video datasets.