Unobtrusive Sensor Fusion Fall Detection Mat Designed For Elderly Individuals

Falls are the most significant cause of injury in people aged 65 and over. Injury caused by falls such as traumatic brain injuries and hip fractures could be life-threatening. Falls within the older population can have the most devastating results. Most of these falls occur in the home. The largest subgroup of falls at home happens in the bathroom when the user is the most vulnerable. The severity and frequency of these falls are a result of moisture and slip conducive floors. Assistive technology has greatly helped older adults live safer autonomous lives. Fall detection has been an extensively studied and funded topic. Unfortunately, most fall detectors on the market are invasive and often require the user to carry or attach a device to their body. Furthermore, the accuracy of these wearable fall detectors is dependent on the devices' positioning. The proposed system, Catch Me If You Bath Mat (CMBM), a non-intrusive bath mat that detects falls, was created to address this issue. The intention of this system is to detect falling, as well as classify walking and sitting. Additionally, the system can distinguish between a human fall and an object fall. A non-slip bath mat in itself is an effective preventative method, absorbing water that causes slippery falls when exiting a shower or bath. The bath mat developed implements sensor fusion consisting of an accelerometer, gyroscope, ceramic piezo sensor, vibration sensor, and an unweighted and weighted piezo film sensor. Multiple machine learning methods were used to find which time series classification algorithm would lead to the most accurate result. CMBM sensors are mounted beneath a standard bath mat, 17 by 24 inches, and detect falling from at least 2 feet from the microcontroller. Using sensor fusion and an Elegoo Nano microcontroller, we collected data from our four activity classes: human falls, object falls, walking, and sitting. After preprocessing the data, we generated and compared multiple models using R's caret package. Human falls can be detected with an accuracy of 96.15%, a sensitivity of 98.3%, and a specificity of 99.28\% using the proposed system. Furthermore, it is capable of accurately predicting all classes with an accuracy of 97.54%.