Fall Detection Using Pressure Sensor Grid With Machine Learning Classifiers
Generally, falling is prevalent in the elderly due to age-related biological changes. Falls can be life-threatening if noticed late. Hence, a fall detection device is developed to monitor the elderly living alone. The primary objective was to design a pressure sensitive floor mat that is capable of...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2020
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Generally, falling is prevalent in the elderly due to age-related biological changes. Falls can be life-threatening if noticed late. Hence, a fall detection device is developed to monitor the elderly living alone. The primary objective was to design a pressure sensitive floor mat that is capable of reading pressure changes due to different postures and translate them to image. The mat works in unison with a fall detection algorithm capable of differentiating different postures from falls. The other crucial objective was to analyse the accuracy of the fall detection algorithm and to reduce false alarm rate. This research is done to create an indoor-friendly device that differs from commercial fall detection devices such as wearables and vision systems. The test subjects for this research were chosen based on their stature. The subjects closely resembled the height and weight of an elderly. No elderly was involved in this research due to safety concerns. The postures performed by the volunteers were sitting, standing, and lying down or falling. These three postures dictate all the common motions of a human and could be used to accurately predict indoor scenarios. SVM, k-NN, and ANN classifiers were chosen for posture data classification. This was done by analysing popular classifiers used in various fall detection devices. The precision of the system in detecting a fall through an SVM classifier is at 89.37%, while other postures such as standing and sitting yield 92.12% and 85.79% respectively. The overall accuracy of the SVM based system is 89.17%. The precision of the system in detecting a fall through a k-NN classifier is at 85.78%, while other postures such as standing and sitting yield 88.89% and 80.41% respectively. The overall accuracy of k-NN based system is 85%. The precision of the system in detecting a fall through an ANN classifier is at 87.56%, while other postures such as standing and sitting yield 91.87% and 81.44% respectively. The overall accuracy of the ANN based system is 87%. It was found that SVM based fall detection algorithm performs better than k-NN and ANN. A new viable device was created. |
|---|