Online System of Foreground Obstacle Removal from On-Vehicle Video Using Plane Structure Constraints

Abstract

In the urban scene modeling, removing foreground obstacles (e.g. pedestrians, telegraph poles, roadside trees, etc.,) is important problem for 3D reconstruction, privacy protection, etc. We propose the method which can automatically remove foreground objects for such problem from the on-vehicle video data. Since an input data is a video stream and an urban scene is mainly composed of planar surfaces the method can effectively separate the objects by using the Spatio-temporal image analysis. To show the effectiveness of the method, several experiments using real data are conducted, which resulted in a successful separation of complicated objects.