Camera Calibration Using Detection and Neural Networks

Several applications use robotic vision, such as a robot navigating through an unknown surrounding, can use vision as main navigate sensor. This paper focuses on studying camera calibration via stereo vision by means of neural network. A neurocalibration method is proposed based on the neural networks ability to learn nonlinear relationship among a two and three dimension coordinate systems and also its information generalization skill. The data used to train neural network mapping are generated from a calibration grid point obtained through the use of Harris edge detection algorithm. The experimental results indicated that the neurocalibration method is feasible and efficient.