A Vision-based approach to fall detection for elderly patients receiving home-based care
Waruguru, Andrew Kinyua
MetadataShow full item record
Falls present one of the unintentional accidents for people in the world. The adverse effects of a fall vary with the nature of the fall and the impact with the ground or object. Essentially, falls rarely occur in the daily activities of healthy individuals. The occurrence results in fatal or non-fatal falls. However, the falls are consequential for the elderly people since they result in future related problems or death. As such, elderly patients require additional attention in the case of fall events. Therefore, to mitigate the effect of a fall on an elderly patient, there must be the provision of a fast response mechanism. Response time to medical emergencies plays a key role in patient survival and recovery. As such, medical personnel strive to reduce the response time. Proper and immediate notification of an emergency aids in reducing the response time. In order to substantially reduce the negative effect of the fall or increase the survival chances, patients ought to receive fast medical response. Therefore, the need of a fast and proper notification method that aims at providing relevant information in regards to the nature of emergency of the patient. As such, proper monitoring leads to a reduced response time. Arguably, elderly patients require urgent medical care in case of a fall. This research work proposes a multi-person fall detection system, which implements a vision-based approach for fall detection leveraging on region-based convolution neural network. A fixed camera serves as the input device to capture images of people. The system analyses the image to identify the posture and orientation of the people present in the image. Based on the provided image, the system then classifies the occurrence as a fall or non-fall using the developed model. If it identifies a fall, an alert is then sent to a concerned party. The system achieves a mean average precision of 0.8 in fall detection. Further, the system detects a fall in an image in 3.8 seconds thus improving the response time of the medical personnel to aid in curbing the negative effects of a fall on a patient.