Estudio e implementación de algoritmos de detección de silueta de una persona en imagen digital

Abstract

In this project a systematic review of the literature on detection techniques people using the methodology proposed by Kitchenham is presented. The overall search process consists of three phases: search planning, conducting search and presentation of results. Therefore, with this search methodology, collection, classification and analysis of information it is able to reduce the number of existing jobs. Twenty-six studies have allowed the selection of the algorithms Support Vector Machine (SVM), AdaBoost and Histograms of Oriented Gradients (HOG). These algorithms will be implemented and it will make use of the library OpenCV computer vision, these algorithms will be tested with images from the public database "Penn-Fudan Database". Finally, to evaluate the results obtained with the proposed algorithms have been used the confusion matrix for SVM-Adaboost-HOG HOG and algorithms. Six metric will be used from that matrix such as accuracy, precision, sensitivity, specificity, detection rate and error rate. The results obtained with the SVM algorithm-HOG are: 0.74, 0.96, 0.66, 0.94, 0.66 and 0.34. While in Adaboost-HOG: 0.72, 0.85, 0.72, 0.71, 0.72 and 0.28, respectively.