Currently, research in face recognition systems mainly utilized deep learning to achieve high accuracy. Using deep learning as the base platform, per frame image processing to detect and recognize faces is computationally expensive, especially for video surveillance systems using large numbers of mounted cameras simultaneously streaming video data to the system. The idea behind this research is that the system does not need to recognize every occurrence of faces in every frame. We used MobileNet SSD to detect the face, Kalman filter to predict face location in the next frame when detection fails, and Hungarian algorithm to maintain the identity of each face. Based on the result, using our algorithm 87.832 face that must be recognized is reduced to only 204 faces, and run at the real-time scenario. This method is proven to be used in surveillance systems by reducing the computational cost.Keywords:Â Hungarian algorithm, Kalman filter, multiple face tracking, video surveillance system.
Copyrights © 2021