Enhancing Occlusion Handling in Real-Time Tracking Systems through Geometric Mapping and 3D Reconstruction Validation

Abstract

Object detection is a classic research problem in the area of Computer Vision. Many smart world applications, like, video surveillance or autonomous navigation systems require a high accuracy in pose detection of objects. One of the main challenges in Object detection is the problem of detecting occluded objects and its respective 3D reconstruction. The focus of this paper is inter-object occlusion where two or more objects being tracked occlude each other. A novel algorithm has been proposed for handling object occlusion by using the technique of geometric matching and its 3D projection obtained. The developed algorithm has been tested using sample data and the results are presented.