Analysis for Symptoms of Human Fall using Pre-Processing and Segmentation based on Deep Learning Architectures

Abstract

By building sensor-based alert systems, physical therapists can not only decrease the after-fall repercussions but even save lives.Older people are prone to several diseases, and falling is a regular occurrence for them during this period.Various fall detection systems have recently been developed, with computer vision-based approaches being one of the most promising and effective. Here, the sensor-based data has been analysed for a patient's human fall symptoms. This data has been pre-processed using Gaussian filtering with kernel neural network in which the data has been normalized and trained based on neural network. The trained normalized data has been segmented using encoded Stacked Deconvolutional Network (EnSt-DeConvNet). We found that the suggested method predicts such fall symptoms with the highest accuracy from sensor data. Other algorithms' accuracy results, on the other hand, are also fairly close. Experiments reveal that the suggested technique, when compared to other generally utilized techniques based on multiple cameras fall dataset, produced reliable findings and that our dataset, which consists of more training samples, produced even better results. Experimental results showaccuracy of 96%, Precision of 94%, Recall of 88% and F-1 score of 82%, computational time of 69%.