Robust Vehicle Detection and Viewpoint Estimation with Soft Discriminative Mixture Model

Chen Tao, Lu Shijian
IEEE Transactions on Circuits and Systems for Video Technology (CSVT)
Publication Date: 
1 Dec 2016
Strategic Thrust: 
Vehicle detection and vehicle viewpoint estimationare crucial for assistive and autonomous driving systems. In thispaper, we propose a soft, discriminative mixture of viewpointmodels (SDMoV) for joint vehicle detection and vehicle viewpointestimation. The proposed SDMoV model is learned in two steps.First, a discriminative viewpoint-specific component model islearned for each cluster of vehicle images with similar viewpoint,which aims to maximize vehicle viewpoint classification accuracy.Second, a new soft margin objective function is designed to retrainthese component models into a mixture of viewpoint models,which aims to maximize the vehicle detection accuracy. Onedistinctive feature of the SDMoV model is that it is capable ofdetecting vehicles and estimating their viewpoints simultaneously.Experiments on three state-of-the-art datasets show that theproposed SDMoV model achieves superior accuracy for bothvehicle detection and vehicle viewpoint estimation tasks.