Performance Evaluation and Scheme Selection of Person Re-Identification Algorithms in Video Surveillance

Ju Liu, Lingchen Gu, Yu Lu, Peng Lan

OAI: oai:igi-global.com:238884 • DOI: 10.4018/IJDCF.2019100104

With the increasing number of camera networks deployed in public places, intelligent video processing has become a key technology for video surveillance. In order to alleviate the workload of the tracers in the artificial tracking video, person re-identification (re-id) can match a large number of pedestrian images to obtain the location of same person at different time in surveillance. This article focuses on the comparison of different classic distance metric learning methods so as to select optimum person re-identification scheme with excellent performance. The authors compare four algorithms matching Local Maximal Occurrence (LOMO) feature representation on three common databases and obtains a criterion to choose algorithms for different datasets. The selection of re-identification algorithms can simplify the video investigation process according to the size and number of person images. In the end, they propose an improved metric learning based on one of algorithms and get improved results. The re-id is useful and efficient in works such as the criminal investigators etc.

Digital Crime & Forensics Security and Forensics IT Security and Ethics