Space location of a point using artificial vision

Abstract

With the purpose of locating a point in the space using WebCams (artificial vision), the following aspects were considered: camera parameters, type of process to carry out on the captured image and presentation of the results. With the objective to find the camera useful field of vision, the angle for which the radial distortion is minimum was experimentally determined; then using a Multilayer Perceptron neural network with 3 inputs, 7 hidden neurons and 10 outputs, the images from each camera are filtered to identify a specific color; averaging the resulting set of points, the center of the object is bidimensionally located in each image, with this information and applying a mathematical development, the center of the object is completely located in a random XYZ orthogonal axis, respect to which, is only needed to know the position of each camera. A computer software, that allows to observe the images of each camera and the configuration of most of parameters, was developed. In the system calibration is necessary to consider: cameras direction, suggested distances between the centers of cameras and quality of the obtained images.