Seeing through wearables: A comprehensive face recognition dataset from body worn cameras

Body worn cameras (BWCs) have become more and more popular over the last decade. They are becoming one of the essential tools for law enforcement officers to carry with them for surveillance purposes. Generally, videos captured by BWCs are used a posteriori through visual inspection in case of major problems between police officers and citizens. Limited academic research has been conducted on image and video processing using BWCs. There are extremely few datasets available that are based on BWCs. For this objective, we introduce FALEBface: a novel dataset for face detection and recognition using BWCs. We also provide some baseline experiments on the proposed dataset. This work includes two distinct insights: (1) introduction of a dataset specific to body cameras with the applications of facial recognition, and (2) evaluation of models in different environments. The experiments are carried out in three environments: indoor, outdoor, and dark which includes a variation of expressions for the subjects, and a comparative study is also done to check the performance of the models across environments and spectra. To facilitate further research in this domain, the entire dataset can be obtained upon request from the authors (The dataset can be obtained by visiting https://faleb.eurecom.fr/).