Global Identification of Tracklets in Video Using Long Range Identity Sensors

Abstract

Reliable tracking of people in video and recovering their identities are of great importance to video analytics applications. For outdoor applications, long range identity sensors such as active RFID can provide good coverage in a large open space, though they only provide coarse location information. We propose a probabilistic approach using noisy inputs from multiple long range identity sensors to globally associate and identify fragmented tracklets generated by video tracking algorithms. We extend a network flow based data association model to recover tracklet identity efficiently. Our approach is evaluated using five minutes of video and active RFID measurements capturing four people wearing RFID tags and a couple of passersby. Simulation is then used to evaluate performance for larger number of targets under different scenarios.

Detecting and identifying people in mobile videos

“…In the experiment, we only use fixed camera tracking to locate people, without actually applying the hybrid RF system to find out targets' unique identities that are required for an application to query target information from a database, to rule out the influence of identification errors evaluated in [8].…”

“…First we use a hybrid video and RF tracking system as the infrastructure to accurately locate the observer and the targets [8]. Fixed mounted cameras can detect human targets and achieve localization accuracy within a meter, but they cannot find out the identities of the targets.…”

A calibration free hybrid RF and video surveillance system for reliable tracking and identification

“…We introduced a network flow based two stage global tracklet identification algorithm, which enables us to use video and active RFID technology to address the above problems [14]. However, this system requires calibration of the camera and the RF devices, and can only work in an environment with no obstacles.…”

“…The variance of RFID measurement noise we use is 5.8dB from field measurements. The RFID reader locations are given to the original algorithm [14], and are kept unknown for the Simultaneous Identification and Mapping algorithm. We run 4 scenarios with different number of targets wearing RFID tags ( ID N ) and number of targets not wearing RFID tags ( NID N ).…”

Tracking by Identification Using Computer Vision and Radio

“…However, contrary to the system we propose, their system is not concerned with localization of individuals; their division of area of interest in relatively crude locations is done solely to allow association of detections from video cameras with those from the RFID system. Yu and Ganz [37] aim to prevent identity switches in longer video sequences by associating identities from a signal-strength-based active RFID system with tracklets that are obtained from a calibrated video camera. Later, they extended their approach to use raw (uncalibrated) radio measurements, and perform tracking and identity association in image plane instead on the ground plane [38].…”