Vehicle-identification Based on Complementarity Feature

Vehicle re-identification is one of the important tasks in the field of intelligent transportation, it is indispensable in intelligent traffic police, intelligent security and other application situation. In this paper, we proposed a novel complementarity feature for re-identification. The algorithm is divided into two stages: online and offline. A view discriminator is built to determine whether the view of each vehicle image is front-view or back-view. Then each vehicle image is assigned a reverse sample based on the vehicle’s view information. In the online phase, we train a Complementarity Feature generator by using the features of the reverse sample as the supervised information. Experimental results show that our algorithm is superior to existing methods.