Detección automática de victima con cuerpo parcialmente obstruido utilizando Kinect.

Abstract

Object detection is a fundamental problem in computer vision, at present there are several techniques to solve this problem. Automatic recognition systems and feature extraction methods are solutions for locating an object in an image. The automatic image analysis uses different sensors to interpret the stage and the features you want to extract. These more known sensors such as cameras, are of different types and for each application. Tracking people is one of the most common applications in machine vision, where there are different methods and algorithms to identify them. These methods and algorithms are more focused on identifying people standing because it is most common in almost all situations, but lying body is not normal to take it to the studio, but in cases of some natural disasters which is vital rescue people or the worst out bodies, these bodies are usually trapped in the dark and lying in rubble. All these aspects are also study where a tool that works in the dark and detect lying bodies, a tool that can help solve this problem is the Kinect is required. The Kinect sensor is a tool of artificial vision that is usually used for video games, more precisely for the Xbox but in recent years a version for computers was launched through the Windows operating system, so that developers can create projects and expand the range of users who intend to reach. Today there is already a database with projects implemented with Kinect which facilitates research and development. The Kinect has an infrared laser projector and a pair of cameras, these captors are CMOS image sensors, one is a color RGB camera and the other is also a camera but this detects reflections of the infrared beam. [1], [2] The algorithms that already brings in programming is the recognition of the skeleton which is the most widely used, but the body of the person must be standing or sitting, this project aims to provide a solution, in a case the body it is not in the default position if you are lying and you can visualize any part of the body, because it may give some scenarios with these features, this application can be helpful in some scenarios of natural disasters. Because of this, it was possible to implement a system to detect a body lying position and partially obstructed, using techniques such as cross-correlation and normalized, providing a solution in this type of scenario.