Global Identification of Tracklets in Video Using Long Range Identity Sensors

Abstract

Reliable tracking of people in video and recovering theiridentities are of great importance to video analytics applications.For outdoor applications, long range identity sensorssuch as active RFID can provide good coverage in alarge open space, though they only provide coarse locationinformation. We propose a probabilistic approach usingnoisy inputs from multiple long range identity sensorsto globally associate and identify fragmented tracklets generatedby video tracking algorithms. We extend a networkflow based data association model to recover tracklet identityefficiently. Our approach is evaluated using five minutesof video and active RFID measurements capturing four peoplewearing RFID tags and a couple of passersby. Simulationis then used to evaluate performance for larger numberof targets under different scenarios.identities are of great importance to video analytics applications.For outdoor applications, long range identity sensorssuch as active RFID can provide good coverage in alarge open space, though they only provide coarse locationinformation. We propose a probabilistic approach usingnoisy inputs from multiple long range identity sensorsto globally associate and identify fragmented tracklets generatedby video tracking algorithms. We extend a networkflow based data association model to recover tracklet identityefficiently. Our approach is evaluated using five minutesof video and active RFID measurements capturing four peoplewearing RFID tags and a couple of passersby. Simulationis then used to evaluate performance for larger numberof targets under different scenarios.