Aiming at the problems of low efficiency and slow speed in current fall detection algorithms, a novel human posture-based fall detection algorithm is proposed. This algorithm obtains human skeletal key points information based on OpenPose and determines the human fall state based on three criteria: the descent speed of the center of gravity, the body tilt angle, and the deformation ratio of the body contour. During the experimental phase, compared with methods solely based on deep learning or wearable devices, the proposed algorithm shows the best performance, with a detection sensitivity of 98.35%, specificity of 96.79%, and accuracy of 97.11%. The experimental results verify the stability and reliability of the proposed algorithm, which has a broad application prospect.