Pedestrian Detection Using Thermal Imaging for Night Driving Assistance

Pedestrian detection is an important problem in the design of driving assistance systems that can reduce accidents and save lives. Although many annotated visible pedestrian datasets are publically available, similar annotated thermal datasets are rare. Such datasets are essential for training a classifier that can be used for pedestrian detection at night from thermal images. Manual annotation for large datasets is a tedious and time consuming option. In this paper, we propose an automatic alternative for constructing an annotated thermal imaging pedestrian dataset by transferring detections from registered visible images simultaneously captured at daytime where pedestrian detection is well developed in visible images. Histogram of Oriented Gradients (HOG) features are extracted from the constructed dataset and then fed to a discriminatively trained part based classifier that can be used to detect pedestrians at night. The resulting classifier was tested for night driving assistance and succeeded to predict pedestrians even in the situations where visible imaging pedestrian detectors failed because of low light or glare of oncoming traffic. KeywordsPedestrian Detection; Thermal Imaging; Driving Assistance