Human detection in near-infrared spectrum
Date of Issue2014
School of Electrical and Electronic Engineering
Fast growing human detection technology has been widely applied in different industries. There is a potential for using near-infrared spectrum to perform the detection task. This dissertation is to study the performance of existing human detection algorithms working in near-infrared spectrum. The human detection task can be accomplished by these two types of detection methods: full body detection and face detection. Several well-known algorithms for detecting full bodies and faces are evaluated based on dataset collected in daytime and nighttime. In daytime, both images in visible spectrum and near-infrared spectrum are collected while in nighttime only near-infrared images are collected. The evaluation of these detection methods involves comparisons of different methods in different spectrums at different time. The comparison results show the potential of using near-infrared spectrum to detect humans. A tool for ground truth annotation is implemented to reduce the workload of the evaluation process. A novel algorithm for bounding rectangle grouping is also implemented to support the detection experiments.
DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing