Single and Multi-Spectral Methods for Pedestrian Detection: A Comprehensive Review

Real-world object identification applications and theoretical research, for both of them, pedestrian detection is still a crucial task. In order to combine classifiers and characterize pedestrian characteristics, traditional pedestrian detection algorithms need specialists to design features. Convolutional neural networks (CNNs) and their evolved forms have become a key component of deep learning. This review presents detailed spectral-based pedestrian detection methods. The review reveals that single spectral deep feature based pedestrian identification system performs better than channel feature based approaches, but for dark environments, pedestrian detection multi-spectral methods and datasets need to be focused. In addition, key accessible data sets and evaluation parameters have been addressed with respect to the pedestrian detection problem. Future developments in pedestrian detection could focus on enhancing the precision and speed of detection under all environmental conditions.