PRIVACY PRESERVING ELDERLY FALL DETECTION USING KINECT DEPTH IMAGES BASED ON DEEP CONVOLUTIONAL NEURAL NETWORKS
European Journal of Molecular & Clinical Medicine,
2020, Volume 7, Issue 3, Pages 5492-5510
AbstractThe increase in mortality rate of the elderly in the recent years is mainly due to the occurrence of falls. Ensuring safely of the elderly is a crucial task since they cannot be monitored constantly all the time. In this research, we propose a novel scheme for the fall detection of the elderly using depth videos acquired from two Kinect sensors. The two sensors include the frontal-mounted Kinect sensor and the depth-mounted Kinect sensor. Fall detection using Kinect depth maps offers a low-cost, reliable and privacy preserving solution. Using the depth maps, depth motion maps (DMM) are generated. From these maps, depth ConvNet features are extracted using deep convolutional neural network (DCNN) structure. In this research, two convolutional layers and two max-pool sub-sampling layers are employed. The obtained features have extremely high discriminative strength to distinguish between fall and non-fall actions. From the extracted features, fall detection is done using Extreme Learning Machine (ELM) classifier. The temporal frame length of the depth motion maps is varied, and the corresponding fall detection accuracy is obtained for the identification of optimal temporal length. Since, this framework utilizes only Kinect depth maps, the privacy issue involved in fall detection during the usage of RGB videos are eradicated. Evaluation was done using publicly available University of Rzeszow Fall Detection (URFD) dataset. Performance evaluation was done based on confusion matrix obtained. Metrics like specificity, precision, sensitivity, and F-score were evaluated from the obtained confusion matrix. Evaluation results clearly show the efficacy of the proposed fall detection framework by comparing with the other state-of-the-art fall detection works proposed in the literature.
- Article View: 63
- PDF Download: 133