Pipelined architecture for real time detection and tracking of moving objects on an hybrid device

In this paper, we present a visual tracking approach based on Dynamic Neural Fields (DNFs) in video images. To perform a task like tracking, a real time object detection application is applied first to identify the most salient regions or objects in input visual scene by optimizing a visual bottom-up attention model. The output saliency map is then used as input for DNFs with its ability to track objects and object tracking results give only the most active peak position that represents the position of the tracked object. To allow this proposed approach to be applied on mobile robots and vehicles, we are motivated to present a new homogeneous multistage architecture that supports the proposed algorithm. More specifically, the parallel pipelined architecture is composed of several computational stages where each one connected in a hypercube topology through a communication links. Consequently, we especially designed and implemented SCM-SCComCM skeleton as an efficient parallel programming model for the proposed application. The experimental results show that the proposed parallel architecture allow us to implement large DNF architectures that can run in real time and be used in real-world applications.