Person tracking with non-overlapping multiple cameras

Sanjay Kumar Sonbhadra, Sonali Agarwal, Muhammad Syafrullah, Krisna Adiyarta

Abstract


Monitoring and tracking of any target in a surveillance system is an important task. When these targets are human then this problem comes under person identification and tracking. At present, large scale smart video surveillance system is an essential component for any commercial or public campus. Since field of view (FOV) of a camera is limited; for large area monitoring, multiple cameras are needed at different locations. This paper proposes a novel model for tracking a person under multiple non-overlapping cameras. It builds the reference signature of the person at the beginning of the tracking system to match with the upcoming signatures captured by other cameras within the specified area of observation with the help of trained support vector machine (SVM) between two cameras. For experiments, wide area re-identification dataset (WARD) and a real-time scenario have been used with color, shape and texture features for person's re-identification.

Keywords


Non-overlapping multiple cameras; Person detection; Person identification; Support vector machine; Video surveillance

Full Text: PDF

Refbacks

  • There are currently no refbacks.