Moving object detection in vehicle mounted dashcam videos

Abstract

Moving object detection has been an active research area for many years. In spite of the advancements made in this area, moving object detection in moving cameras still remains a challenging problem due to the introduction of camera motion along with the motion of the object. In this work, we propose a CNN based method to detect moving objects in dashcam videos. The feature which we use to identify the moving objects is optical flow. First, we use the Flownet to extract the optical flow RGB images from two consecutive video frames. We then feed the obtained optical flow based RGB image to Faster-RCNN model. This model then identifies the target objects in the video. We have used two datasets, namely KITTI dataset and BDD-mod dataset, for the evaluation of the method.