Performance analyses and optimization of Real-time Multi-step GA for visual-servoing based underwater vehicle

Genetic algorithm (GA) has been applied for real-time pose estimation in this research because of its simplicity, global perspective and repeatable ability. In many of these different situations or problems, optimum selection parameters are a critical factors in the performance of GA. We have developed visual-servo type underwater vehicle using dual-eye camera and 3D marker using real-time pose tracking, named as Real-time Multi-step GA. The relative pose between a vehicle and a 3D-marker can be estimated by Model-based matching method. To recognize the pose of the marker with respect to the vehicle, it is needed to utilize the optimum searching in real-time, and the real-time pose estimation problem can be converted into an optimization problem over a time-varying distribution function with multiple variables. Therefore, analyses the convergence performance of real-time multi-step GA for 3D model-based recognition for underwater vehicle was conducted and reported in this paper. The main aim of this paper is to choose the best parameters for GA that are optimized over population size, selection rate, mutation rate based on their relative fitness value to improve the performance of searching in time domain. The experimental results show that the proposed system effectively improved the searching performance of Real-time multi-step GA for real time pose tracking, having enable an automatic docking of underwater vehicle by dual-eyes visual servoing.