Real-Time Detection of Pedestrians: A Comparison of Three Segmentation Algorithms in Infrared Video

Real-Time Detection of Pedestrians: A Comparison of Three Segmentation Algorithms in Infrared Video

Juan Serrano-Cuerda (Universidad de Castilla – La Mancha, Spain), José Carlos Castillo (Universidad Carlos III de Madrid, Spain), María T. López (Universidad de Castilla – La Mancha, Spain) and Antonio Fernández-Caballero (Universidad de Castilla – La Mancha, Spain)
Copyright: © 2016 |Pages: 19
DOI: 10.4018/978-1-5225-0245-6.ch013
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

Real-time pedestrian detection is a key technology for video surveillance. A widespread approach for detecting pedestrians is the use of color information. In recent times, the use of thermal infrared cameras has revealed to be an excellent alternative that offers good results in people segmentation. Nonetheless, thermal infrared cameras are very sensitive to the overall heat detected at each image. Moreover, a great amount of infrared images has low spatial resolution and lower sensitivity than visible spectrum images due to the technological limitations of infrared cameras. This chapter introduces a comparison of three different algorithms for real-time and robust pedestrian detection in the infrared spectrum. The aim of the paper is to look for the best algorithms prepared to resolve the conflicts that arise in the detection process in image sequences. We propose to use simple rules as conflict resolution mechanism when the outputs of the three algorithms do not coincide.
Chapter Preview
Top

Background

Detecting pedestrians is a key technology for many applications, especially in the video surveillance domain (Dollár, Wojek, Schiele, & Perona, 2012). At the same time, it is one of the most challenging problems in computer vision and remains a scientific challenge for realistic and dynamic scenes. Indeed, visual processing of pedestrians, including detection, tracking, recognition, and behavior interpretation, is a key component of intelligent video surveillance systems. A number of surveillance applications require the detection and tracking of people to ensure security and safety (Navarro, Fernández-Caballero, & Martínez-Tomás, 2014), (Costa, Guedes, Vasques, & Portugal, 2013). That is, many video surveillance systems require the ability to determine if an image region contains pedestrians. This is none but a specific case of object classification in which there are only two object classes: pedestrian and non-pedestrian. Object classification in general is difficult and people detection is even harder. In addition, video-surveillance systems must run at video-rate and thus require a trade-off between precision and computing time. Moreover, any pedestrian detection method highly depends on segmentation, which remains a primitive problem. A widespread approach for detecting pedestrians is the use of gray scale (Enzweiler & Gavrila, 2009) and color information (Wan & Liu, 2009; Rodriguez & Shah, 2007), (Schwartz, Kembhavi, Harwood, & Davis, 2009). These are usually problematic when facing changes in lighting in a scene or visibility problems therein. To guard against these failures, you can find an alternative in the use of the infrared spectrum (Sun & Park, 2007).

Complete Chapter List

Search this Book:
Reset