Fall Detection Based on HOG Features for Millimeter Wave Radar

The secondary injury caused by falls poses a serious threat to the physical and mental health of the elderly. Millimeter-wave radar has the characteristics of all-weather, unaffected by light, privacy-friendly, and strong penetration. It has the ability to compete with wearable sensors and video monitoring systems. With its excellent performance, it has become a fall detection method that has attracted much attention. In this paper, we propose a novel human fall detection method for millimeter wave radar based on histogram of oriented gradient (HOG) features and machine learning. The manual extracted HOG features of time-range maps, time-velocity maps and time-angle maps are used as the input of support vector machine (SVM) classifier. We separately verify the accuracy of the SVM classifier with single HOG feature and multiple HOG features. Our experiments proved that the accuracy of fall detection can reach 95% by using the HOG features of three maps as the input of the SVM classifier, which is at least 12.5% higher than that of single map.