Enhancement of Missing Face Prediction Algorithm with Kalman Filter and DCF-CSR

Abstract

Detection and tracking of moving objects in sequence videos has wide applications in security surveillance, and becomes concern to many researchers. In actual environmental conditions, with various lighting conditions, object tracking faces a number of challenges including partial or severe occlusion which causes some systems to lose information so that it is difficult to estimate object trajectory. In the domain of surveillance, human tracking should not only be based on face, but also based on other characteristics, so that wherever the person facing towards, the system is always able to do the tracking correctly. In this study the detection and alignment process employed Multi-task Cascaded Convolutional Networks and Kalman Filters to predict facial position. Then, at the times the face is not facing towards the camera, the system saves the color of the bounding box that was last seen and tracks by color using the Discriminative Correlation Filter with Channel and Spatial Reliability (DCF-CSR). The proposed method resulting in increasing a person's detection rate when facing away from the camera.