Improved Post-Processing for Human Detection in Railroad Surveillance

Abstract

Foreground detection (FD) has been widely used for moving object detection such as human detection. The practical performance of FD methods varies significantly in different real-world environments. Post-processing methods can improve the effectiveness of FD algorithms by providing high quality foreground masks for further detection. The accuracy of human detection in railroad surveillance is often limited due to occlusion and noise, causing incomplete objects in foreground masks. In this work, we introduce a novel integration step into the post-processing module following FD. We take priori knowledge into account and apply specific rules when combining the labeled components, resulting in significant improvement in the accuracy of human detection.