BinaryCoP: Binary Neural Network-based COVID-19 Face-Mask Wear and Positioning Predictor on Edge Devices

Face masks have long been used in many areas of everyday life to protect\nagainst the inhalation of hazardous fumes and particles. They also offer an\neffective solution in healthcare for bi-directional protection against\nair-borne diseases. Wearing and positioning the mask correctly is essential for\nits function. Convolutional neural networks (CNNs) offer an excellent solution\nfor face recognition and classification of correct mask wearing and\npositioning. In the context of the ongoing COVID-19 pandemic, such algorithms\ncan be used at entrances to corporate buildings, airports, shopping areas, and\nother indoor locations, to mitigate the spread of the virus. These application\nscenarios impose major challenges to the underlying compute platform. The\ninference hardware must be cheap, small and energy efficient, while providing\nsufficient memory and compute power to execute accurate CNNs at a reasonably\nlow latency. To maintain data privacy of the public, all processing must remain\non the edge-device, without any communication with cloud servers. To address\nthese challenges, we present a low-power binary neural network classifier for\ncorrect facial-mask wear and positioning. The classification task is\nimplemented on an embedded FPGA, performing high-throughput binary operations.\nClassification can take place at up to ~6400 frames-per-second, easily enabling\nmulti-camera, speed-gate settings or statistics collection in crowd settings.\nWhen deployed on a single entrance or gate, the idle power consumption is\nreduced to 1.6W, improving the battery-life of the device. We achieve an\naccuracy of up to 98% for four wearing positions of the MaskedFace-Net dataset.\nTo maintain equivalent classification accuracy for all face structures,\nskin-tones, hair types, and mask types, the algorithms are tested for their\nability to generalize the relevant features over all subjects using the\nGrad-CAM approach.\n