Abstract
The aim of this study is to track the faces of all pedestrians in the video surveillance. Face images from each camera can use Chinese Whisper face clustering algorithm to cluster the same person’s face together, and according to the results of face clustering to find out which people through the camera. Double Triplet Networks (DTN) designed in this study is used to learn the depth features of human face. DTN is trained on LFW data set, and the model trained can improve its recognition accuracy to 99.51% by Margin Sample Mining Loss (MSML) and Focal Loss hard sample equalization. Comparing the similarity of the facial features in same video surveillance areas can track the faces of pedestrians, and comparing the similarity of the facial features in different video surveillance areas can predict which camera area the face comes from and tracking the sequential paths of pedestrians through these areas. Cross-camera face tracking is possible by transmitting facial features between cameras in real-time.
Highlights
At present, deep learning is applied to face recognition, pedestrian tracking, behavior analysis, traffic statistics and action recognition
Cross-camera face tracking plays an important role in the field of surveillance
SYSTEM DESIGN This paper proposes an improved Double Triplet Networks (DTN)
Summary
This work was supported in part by the National Natural Science Foundation of China(NSFC) under Grant 61179019, in part by the National Natural Science Foundation of China(NSFC) under Grant 81571753, in part by the Youth Innovation Talent Project of Baotou under Grant 0701011904
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have