Laser & Optoelectronics Progress, Volume. 59, Issue 2, 0210004(2022)
Multi Face Real-Time Tracking System Based on DTN in Multi Camera Field of View
In order to realize multi face image tracking across camera regions,a cross camera tracking network based on double three branch twin network (DTN) is proposed. The specific method is to apply Chinese Whisper(CW) face clustering algorithm to cluster the face images of the same pedestrian, and determine the captured target face through intelligent monitoring according to the results of face clustering. By improving the network structure and training function of FaceNet, pedestrian face tracking is realized accurately. After training DTN on LFW data set, the face recognition rate can be improved to 99.51% through margin sample mining loss (MSML) and focus loss difficult sample balance training. Experimental results show that by comparing the similarity of face features in the same video surveillance field, the proposed network can track pedestrian face targets through this area; through the real-time transmission of face features between cameras, cross camera face tracking is realized.
Get Citation
Copy Citation Text
Guoyin Ren, Lü Xiaoqi, Yuhao Li. Multi Face Real-Time Tracking System Based on DTN in Multi Camera Field of View[J]. Laser & Optoelectronics Progress, 2022, 59(2): 0210004
Category: Image Processing
Received: Jan. 8, 2021
Accepted: Mar. 9, 2021
Published Online: Dec. 23, 2021
The Author Email: Xiaoqi Lü (635302395@qq.com)