A Robust Saliency Integrated Method for Monocular Motion Estimation

Abstract

The moving objects in the background bring the challenges for motion estimation in the visual odometer applications for autonomous driving. This paper proposed a robust motion estimation method with saliency based outlier removal with a single camera in visual odometer's applications. First, an efficient saliency calculation method is proposed to detect the moving objects from the noisy monocular image sequences. Then the robust motion estimation is conducted with weighted saliency map while considering of textures area and smooth area. Experiments on KITTI dataset are conducted and the experimental results show that our proposed method with a single camera is comparable with the start-of-art stereo motion estimation method and outperform other compared stereo motion estimation methods.