Algorithm for Detecting Human Fall Behavior Based on Pose Estimation

Aiming at the problems of fall detection algorithm such as large amount of calculation, poor real-time performance and difficulty in feature extraction, an improved fall detection algorithm based on human skeleton key point detection and multi-feature fusion was proposed. First, the main network of OpenPose is replaced with MobileNetV3 network, the size of the convolution kernel is adjusted from 7×7 to 3×3. Then, the multiple fall behavior features were fused, and the detection results were output by SVM classification. On the fall dataset, the proposed algorithm achieves high accuracy.