Segmentación de imágenes obtenidas a través de un sensor Kinect con criterios morfológicos y atributos visuales de profundidad

The digital image processing has grown from an assistant to a vital tool in many industrial processes. Technological development has enabled the creation of vision equipment to measure the temperature of a system, others which can capture images at high speeds, and finally those who can quantify the depth at you will find different objects in a scene, among other. This latter uses a technology based in infrarred light emitters and sensors, where this reflected light in the objects allow to make an estimation of objects depth in milimeters at which they are. One of the problems inherent to these devices is the high susceptibility to noise with which results in loss of information. The Kinect sensor is an example of this type of devices and has come to mean a revolution for the features it has and especially for its low cost. This thesis presents a methodology to solve the problem of noise and holes present in depth information delivered by the Kinect using morphological and statistical filters, simultaneously we perform image segmentation by distances in depth and color images comparing it to traditional segmentation processing analyzing their advantages and disadvantages. Visual attributes are used to represent the depth information delivered by the sensor and a technique of projective geometry is applied to tie the depth and color information that sensor delivers, all this in order to bring the same functionality to this low cost device to those that offer industrial grade equipment. As a particular case of study, Kinect sensor and its articulation user detection function was used as auxiliary tool for physiotherapy people at the University to make postural analisys