Análisis de dos algoritmos para detección de rostros e implementación de uno de ellos utilizado en analítica de video para sistemas de videovigilancia

Abstract

The implementation of video surveillance systems with video analysis provides better recording, storage and monitoring features in many deployment environments, allowing the monitoring of objects and people to prevent and guarantee security. At present the most used video surveillance systems are those with IP technology. Face detection is an application of video surveillance systems and is the first step for further applications such as tracing monitoring and recognition. For the facial detection, there are a series of algorithms that allow the extraction of the face in a video image, among which are the cascade method of Viola & Jones and the method by geometric models using Hausdorff distance. The most common problems in the detection of faces that occur in the process of acquiring the images in a video surveillance system are the conditions of lightness, low contrast, foreign objects in the background, partially captured faces, position of the person, Facial expression and image distortion; Problems that are solved if the implementation environment is controlled obtaining the best results of detection rates of faces above 90%. The Viola & Jones algorithm divides the integral image into subregions of different sizes and uses a series of classifiers. The Hausdorff method is based on edges between a general face model. For its reliability and efficiency the algorithm most used for face detection is the method developed by Viola & Jones.