The Study of Pedestrian Re-Identification with the Illumination Change
Chunfang Zhao, Xianqiao Chen · 2018
Pedestrian re-identification has caused the extensive concern of many scholars, but due to the problems such as illumination, pedestrian posture change and keep out, there is a lot of room for the study of pedestrian re-identification. Based on the influence of illumination change on pedestrian recognition, a new method of pedestrian re-identification which adapt to the change of light is proposed in this paper. Firstly, a candidate set of pedestrian features is established, and the local characteristics of pedestrian images are extracted based on the Bag-of-words model. Secondly, a number of sample subsets are obtained to simulate a variety of illumination scenarios. Based on the KISSME algorithm, the measurement matrix of single-illumination scene and cross-illumination scene was obtained. According to the scene, the corresponding measure matrix is selected to represent the similarity between pedestrians. Finally, a comprehensive decision scheme based on multi-feature recognition results is established to obtain the final result of pedestrian recognition. It is proved by experiment that the method of this paper can have a good recognition effect in the environment with the change of illumination.