We present a novel radar-based system for real-time indoor positioning and detection of objects and human-bodies with low-quality, inexpensive sensors. Using modern deep learning methods, we avoid the use of expensive hardware and computationally-expensive signal processing methods for object detection.We train our model on mini-Doppler maps, collected via software defined radios. Crucially, our system is different from existing RF-based detection systems as it operates in a less crowded frequency range of 433 MHz, allowing us to use inexpensive off-the shelf hardware. Our system, based on the VGG-16 model, reports high-accuracy results on: (1) classification of different objects/materials (plastic, glass, metal); (2) detection and classification of multiple visually and materially similar objects and the human-body; and (3) Simple object detection at different distances between the transmitter T x and the receiver R x .WALDO, using low frequency radio waves, is able to handle occlusions and bad lighting environments. Our results demonstrate that Deep Learning methods can be combined with inexpensive, low-frequency radars to achieve high accuracy in real-time on various useful tasks.
