Research Information System
2nd International Conference on Machine Learning and Autonomous Systems, ICMLAS 2025, Bangkok, Thailand, 10 - 12 March 2025, pp.12-18, (Full Text)
Falls represent a considerable risk to the well-being of geriatric populations, frequently resulting in injuries or mortality. In response to this pressing concern, we introduce an innovative fall detection apparatus employing acoustic sensors and FPGA (Field Programmable Gate Array) technology to effectively alleviate these hazards. Our system capitalizes on acoustic phenomena, particularly those associated with footstep sounds and falls, which are gathered through routine activities and video documentation to develop a comprehensive model adept at differentiating between fall incidents and non-fall occurrences. The primary attributes encompass the derivation of Mel-frequency cepstral coefficients (MFCC) from the processed auditory signals. To enhance operational efficacy, an FPGA is utilized to execute all data processing functions, inclusive of Fourier transformations and a threshold-based predictive algorithm. This methodology diminishes dependence on conventional computational systems while preserving elevated processing efficiency. The derived features are subsequently employed to formulate a data classification model utilizing a Convolutional Neural Network (CNN) and Random Forest Tree (RFT), which proficiently discriminates between sounds indicative of falls and those not related to falls.