Human Detection Based on Brightness-histograms Using a Thermal Camera in Forestry Environments

It is essential to have a reliable system to detect humans in forestry environments in close range of forestry machines to stop cutting or carrying operations to prohibit any harm to humans. We use a thermal camera due to its advantages over RGB cameras by considering different lighting and occlusion conditions. This paper introduces a novel shape-independent feature called brightness-histogram for the purpose of human detection. This feature is a histogram that is created based on the pixel values (temperature) of regions of interest in thermal images. In the proposed algorithm, brightness-histograms of human and non-human objects from thermal images are extracted and their similarities is measured by using Euclidean distance. Then a K-Nearest Neighbors (KNN) classifier is trained and evaluated based on similarity values using Euclidean distance to measure the performance of brightness-histograms in human detection in forestry environments. Using this feature we reach to precision rate of 81%, false discovery of 19% and recall of 86%. We also provides a set of practical aspects that are needed to be taken into account when using thermal imaging for human detection in real-world scenarios.