Multiple Moving Targets Tracking from Sensor Scheduling Perspective Chen Wang College of Electrical Engineering Zhejiang University Hangzhou, China 310027 Email: wangchen0511@gmail.com Zhiyun Lin* College of Electrical Engineering Zhejiang University Hangzhou, China 310027 Email: linz@zju.edu.cn Gangfeng Yan College of Electrical Engineering Zhejiang University Hangzhou, China 310027 Email: ygf@zju.edu.cn Abstract—This paper investigates the problem of tracking multiple moving targets by an ultrasonic location system with an active architecture. It is assumed that several moving targets are on the plane and may interfere with each other if they chirp simultaneously. Extended Kalman filters to estimate the positions of all moving targets are run in a centralized computer based on intermittently available distance measurement from the reference nodes. A myopic scheduling scheme together with a grouping approach are proposed to improve tracking accuracy and avoid interference. Two metric functions are used in myopic scheduling, namely, MEE and MAED. The simulations show that MEE may cause over-partiality which leads to bad tracking accuracy for some targets, and MAED can effectively eliminate such over-partiality. Myopic scheduling schemes with both MEE and MAED are realized on the Cricket platform based on active architecture. The experiment results show that active architecture results in better tracking performance than passive architecture for tracking a small number of moving targets. Also, it is shown that in active architecture myopic schemes with both MEE and MAED have better tracking performance than a periodic scheme. I. I NTRODUCTION High tracking accuracy and good real-time performance are expected for a tracking system. In the past several years several methods for indoor localization appeared to achieve these goals. Priyantha [6] designed a Cricket Location Sys- tem. Passive architecture is adopted in the system where the reference nodes fixed on ceiling chirp (sending ultrasonic and RF signals) and the moving targets receive these chirps. It provides high accuracy of position estimation about 10 in an indoor environment when the targets are static. However it is initially designed for static targets localization rather than for moving target tracking. So the localization system uses 1HZ data rate which is not sufficient for moving target tracking. To deal with the deficiency, Nam et al. [5] introduced a scheme for scheduling chirps and data transmission based on the Cricket platform. This scheme significantly improves the real-time performance and the data sampling rate reaches about 20Hz. However Cricket’s passive architecture possesses an in- herent shortcoming in tracking moving targets. Because at each time slot the target measures distance from only one *Corresponding author. Tel: +86-571-87951637; Fax: +86-571-87952152. This paper was sponsored by the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry of China. reference node and the target is moving, the distances to different reference nodes are not measured from the same position, which influences tracking accuracy. But it can be solved by using active architecture, where the moving target actively chirps and the reference nodes fixed on the ceiling hear the chirp simultaneously. Because ultrasonic and RF signals are sent by the moving target, the target’s distances to these reference nodes are obtained from the same position. Smith et al. [8] compared the performance of the two architectures when considering to track one single moving target. The result shows that the active architecture is better than the passive one in tracking precision. Nowadays there are some systems based on the active architecture [2], [4] and [9]. Among these systems, the Active Bats System [4] also uses the ultrasound to measure distances as Cricket. Although using active architecture can significantly improve the performance of tracking a single moving target, there also exists a new drawback. When multiple targets are tracked and they are close to each other, ultrasonic and RF signals from different targets may interfere with each other. In the Active Bats System, the time is divided into time slots and at each time slot there is only one bat allowed to send ultrasonic and RF signals. This method is of low efficiency because some bats are far away from each other and they can chirps simultaneously without interference. Woodman and Harle [10] gave an algorithm to judge which bats do not cause interference. By allowing the bats that do not cause interference chirp simultaneously, they significantly improve the data sampling rate. Along this direction, two problems need to be addressed. The first one is the scheduling problem. That is, how do we schedule time slots to different targets in an optimal way when interference exists between them? The second one is whether active architecture can have better tracking performance than the passive one in tracking multiple moving targets when interference exists. Scheduling problem has been discussed in many applica- tions. Chhetri [3] introduced myopic and non-myopic schedul- ing methods for the situations where different kinds of sensors were used to track one target and at each time slot only one sensor could be used for energy saving. Shi et al [7] discussed the scheduling problem in sensor networks where 59 978-1-4577-1826-7/11/$26.00 ©2011 IEEE ICON 2011