Object setection and recognition in infrared images

Abstract

In this report, we present a series of object recognition and detection methods on infrared images for surveillance applications. The techniques employed are AdaBoost cascade classification method and support vector machine using histogram of orientation gradients feature descriptors. First, we collect a set of infrared spectrum images and evaluate the performance of each method. After analyzing a preliminary test using the infrared dataset, we tune the classification criteria and provide suggestions to improve classification and human detection accuracy. Our experiments results show some good overall improvements with better accuracy and lower error rate.