Abstract

Short range radars can provide robust information about their surroundings under atmospheric disturbances, such as dust, rain, and snow, conditions under which most other sensing technologies fail. However, this information is corrupted by received power noise, resulting in false alarms, missed detections, and range/bearing uncertainty. The reduction of radar image noise, for human interpretation, as well as the optimal, automatic detection of objects, has been a focus of radar processing algorithms for many years. This paper combines the qualities of the well established binary integration detection method, which manipulates multiple images to improve detection within a static scene, and the noise reduction method of power spectral subtraction. The binary integration method is able to process multiple radar images to provide probability of detection estimates, which accompany each power value received by the radar. The spectral subtraction method then utilizes these probabilities of detection to form an adaptive estimate of the received noise power. This noise power is subtracted from the received power signals, to yield reduced noise radar images. These are compared with state-of-the-art noise reduction methods based on the Wiener filter and wavelet denoising techniques. The presented method exhibits a lower computational complexity than the benchmark approaches and achieves a higher reduction in the noise level. All of the methods are applied to real radar data obtained from a 94-GHz millimetre wave FMCW 2D scanning radar and to synthetic aperture radar data obtained from a publicly available data set.