A Comparative Study of Object Detection and Pose Detection for Fall Detection using Detectron2

Falls have a significant impact on mortality, health, and mobility in older persons. The elderly are particularly vulnerable to falls as they can have a very negative impact on their physical and emotional well-being or, in the worst case, result in death. However, with the right technical solutions, the adverse effects of falls can be lessened. These algorithms may be used to enhance lightweight gadgets that can then respond to the requirements of the users, such as notifying carers or emergency services. To enhance the detection rate, it is necessary to identify and analyze various types of postures effectively. The system is supposed to handle situations that are unclear, such as different body postures, viewpoints, and environmental shifts. This study aims to analyze vision-based complex key point analysis and posture estimation models for object detection and anomalous activity recognition (fall detection). A comparative analysis is done in this paper for both the computer vision techniques using Detectron2 for benchmark fall detection dataset (CAUCA Fall Dataset) in order to analyze the impact of bounding box technique and key point feature analysis for effective fall detection system.