Distributed Sensor Networks for Scene Analysis in GPS-Denied Environments

Sensor parameter estimation is a key process that must be considered when performing data fusion in a multi-sensor object tracking scenario. For example, significant relative time delays in sensor data arriving at a fusion centre can result in a reduction of track accuracy, false tracks, or early termination of a true object track. The same issues may arise in the presence of some relative angular bias between sensors.

We have developed a technique for simultaneous target tracking and estimation of relative time delays and angular biases in data for a multi-sensor system with no access to a Global Navigation Satellite System. The technique makes use of a hierarchical Bayesian model and couples a particle method with an array of augmented state Kalman filters to accomplish this.

The method is used for spatio-temporal bias correction. Initial results demonstrate a significant improvement in tracking performance when registration errors are corrected with the proposed method, as well as a noticeable increase in accuracy over object tracking with only a single sensor. This highlights the benefit of tracking with multiple sensors and the importance of calibration.