Evaluation of thermal imaging for people detection in outdoor scenarios

In this work we evaluate the performance of a trained people detector like Implicit Shape Models (ISM) on thermal images. We compare its performance against two baseline algorithms. One is a simple thresholding approach that detects high temperature blobs in the same thermal images, an approach which is commonly used. Additionally, we compare the results on thermal data to results of the same detector working on images from the visible spectrum that are taken from a camera mounted close to the thermal one. Experiments are conducted on both indoor and outdoor data. Outdoor data was recorded at different temperatures and with a moving robot. The conclusion of this paper is that for indoor usage under normal circumstances simple blob detectors are sufficient. But outdoors a trained people detector outperforms the blob detector by far. Also, the results of the blob detector get much worse on warmer days which hints in the direction that blob detectors will also fail during unusual circumstances indoors, like during a fire. The use of thermal images helps the ISM detector to distinguish people from background clutter and, therefore, the performance of the detector is better on thermal images than on images from the visible spectrum showing the same scene.