Algorithm for Person Tracking on Video Sequences Using Face Identification for Indoor Surveillance

Abstract

This paper discusses the algorithmic framework for tracking people on indoor video. To improve tracking accuracy was used face identification algorithm to reduce errorr rate during complicated trajectory of persons in indoor environment. Object detection was performed with CNN Yolov3 that extract rectangular area as a result. Face detection task was resolved eith Cascade CNN MTCNN with following recognition using CNN MobileFaceNetwork. To form person features we used historgrams in HSV colorspave and CNN that includes 29 convolution layers followed by fully connected layer. The Hungarian algorithm was used as decision maker for allignment problem. Experiments were conducted on five videosequences with the variable number of people in it. The main characteristics of the developed algorithm are obtained which confirmed its effectiveness and the possibility of use for indoor video surveillance.