LOWNET: PRIVACY PRESERVED ULTRA-LOW RESOLUTION POSTURE IMAGE CLASSIFICATION

Indoor posture recognition is vital for monitoring/detecting exercises, activities of daily living, accidental falls, unusual behavior, etc. However, high-resolution image based systems have a high accuracy, they are considered as intrusive and most of the current state-of-the-art image classifiers (VGG, ImageNet, ResNext) are not applicable for ultra-low resolution (<32 pixels in extent) image classification due to their downsizing feature extraction architecture. Thus, we propose a shallow LowNet model for classifiyng privacy preserved 8×8 posture images with its feature preserving architecture, variable ReLU slopes, and custom loss function. LowNet outperformed the existing models (LeNet, ResNet-1, ResNet-2, CNN) on our ultra-low thermal posture (ULTP-51) dataset (available here) consisting of 5843 samples of 51 postures performed by 23 volunteers. Moreover, the positive impact of variable ReLU slopes and custom loss function are reported. Thus, we conclude that LowNet has a great potential in the privacy-preserving device-free sensing technology.