Object Detection of Omnidirectional Vision Using PSO-Neural Network for Soccer Robot

The vision system in soccer robot is needed to recognize the object around the robot environment. Omnidirectional vision system has been widely developed to find the object such as a ball, goalpost, and the white line in a field and recognized the distance and an angle between the object and robot. The most challenging in develop Omni-vision system is image distortion resulting from spherical mirror or lenses. This paper presents an efficient Omni-vision system using spherical lenses for real-time object detection. Aiming to overcome the image distortion and computation complexity, the distance calculation between object and robot from the spherical image is modeled using the neural network with optimized by particle swarm optimization. The experimental result shows the effectiveness of our development in the term of accuracy and processing time.