Vehicle Re-Identification Based on Coupled Dictionary Learning
Panpan Wang, Bo Li · 2018
Vehicle re-identification technology is a key technology to identify unlicensed vehicles in the cross-view of video. Existing vehicle re-identification methods often focuses on the extraction of vehicle license information. There is not a good way to re-identify a vehicle without a license plate or a vehicle with a blocked license plate. In this paper, a method based on feature fusion and coupled dictionary learning is proposed for vehicle re-identification in the absence of a license plate or license block. In dictionary learning, a new least squares coupled dictionary learning method (nLSCDL) is proposed. We design and join the label matrix and mapping matrix to reduce the intra-class disparity of the same vehicle and widen the inter-class disparity of different vehicles across the field of vision. The method has good robustness to illumination change, perspective change and the parameters of the camera itself. our method has good recognition rate under three different vehicle data sets and GRID data sets.