Resource-aware architectures for particle filter based visual target tracking

There are a growing number of visual tracking applications for mobile devices such as smart phones and smart cameras. However, existing computer vision algorithms are demanding and mobile devices possess limited computational capabilities, energy, and bandwidth to support them. Conventional approaches to distributed target tracking with a camera node and a receiver node are either sender based or receiver based. Both approaches are highly suited for certain scenarios, but have limited applicability outside of their scope. In this paper, we propose two new approaches for a particle filter based tracking system. The first proposed approach reduces the energy and bandwidth typically required for the receiver based setup. The second proposed approach partitions tracking workload between sender and receiver and adapts to the frame-to-frame demands of particle filtering. In doing so, this scheme promotes better balance of computing capabilities, energy, and bandwidth among sender and receiver. Results show that the proposed solutions require low additional overhead, can improve on tracking system lifetime, and may be more effective than conventional architectures for many tracking instances.