Human fall detection in infrared images is not affected by ambient light and has important research and application value in intelligent security. Existing fall detection methods do not fully consider the position change law of key points on the human body, which can easily cause false detections of similar fall movements. To solve this problem, we propose an infrared image fall detection algorithm based on an improved alpha pose. The algorithm uses the YOLO v5s object detection network to directly classify human poses while extracting the human body target frame and inputting the pose estimation network. It then evaluates it in combination with the position information and posture characteristics of the key points of the human skeleton. Experiments showed that the algorithm exhibited good performance in terms of accuracy and real-time performance.