An embedded architecture for real-time object detection in digital images based on niching particle swarm optimization

Particle swarm optimization (PSO) is an efficient population-based optimization technique. Niching is introduced into PSO to deal with multimodal problems, and converge onto the largest possible number of minima. In this paper, a niching PSO algorithm is described, which targets the imaging domain; our niching PSO enhances the fitness of a particle dynamically, based on the current position of the particle’s neighbors, while also contemplating the chance of stopping exploration by halting a particle over favorable regions of the fitness landscape. It also caters for repulsion among closely located neighbors, which induces the particles to spread over the region of interest instead of converging onto a single punctual optimum. Based on this technique, an architecture is proposed and implemented for detecting objects in digital images. The new architecture, being efficient and portable, can be used in embedded computer vision and pattern recognition applications. We have assessed the architecture performance in the real-world task of detecting license plates in digital images, an instance of the more general class of object detection problems for which it has been designed. In doing so, we have compared our results with respect to both a corresponding sequential implementation and a standard computer implementation of a classical algorithm which solves the same problem.