A comprehensive comparison of accuracy and practicality of different types of algorithms for pre-impact fall detection using both young and old adults

Analyzing and Comparing Deep Learning Models on an ARM 32 Bits Microcontroller for Pre-Impact Fall Detection

“…Typically, CNN-LSTM architectures consist of three convolutional blocks and two LSTM blocks. For instance, in the work by [4], this architecture achieves an impressive 99.16% accuracy, 99.32% sensitivity, and 99.01% specificity using a 50-step input window with 9 input characteristics. Similarly, [25] reports an accuracy of 97.52% using the same type of model with a sliding window size of 200 steps for 6 input features.…”

“…One major hurdle to the deployment of wearable sensors on elderly is the accuracy of pre-impact fall detection system, often generating false alarms,i.e., alerting fall during activities of daily living (ADL). Thus, most of recent researches have focused on reducing false alarms and improving accuracy of the preimpact fall detection system [2]- [4] using either thresholdbased algorithm [5]- [18], conventional machine learning (ML) [15], [19], or deep learning approaches [20]- [36]. Thresholdbased algorithms have demonstrated good performances to distinguish pre-impact falls from ADL: sensitivities are in the range of 92% -99%, specificities are in the range of 95% -98% with an average lead time (time interval between fall detection and body-ground impact occurrences) in the range of 250 -400 ms. Two interesting papers present exceptional results with a sensitivity and specificity > 97% [17], [18] but the algorithms were evaluated on a very small sample of people.…”

“…This comprehensive review of existing research highlights the absence of studies developing pre-impact fall detection systems using deep-learning models integrated into microcontrollers for elderly care. Notably, the work by Yu et al [4] is the only one that employs an embedded micro-computer (Raspberry Pi 4) to measure lead time and computation latency on a test set.…”

Pre-Impact and Impact Fall Detection Based on a Multimodal Sensor Using a Deep Residual Network

“…Conventional pattern classification and recognition based on DL are the two primary categories of ML techniques [15]. Standard recognition techniques (such as the support vector machines (SVM) technique [16] and the k-nearest neighbor (KNN) technique [17]) would depend on manually extracted features for identification.…”

“…There were three reasons about the selection of the window size of 0.5 s and the overlapping proportion of 50% as follows: (1) the 0.5 s window size in the sliding windows were sufficient to achieve the high detection rates for the fall detection problems as suggested by Liu et al [24]. (2) Since the focus of this study was pre-impact fall detection, the window size was suggested for the fall injury prevention based on the practicality measures [15]. The sliding window approach was the most widely employed segmentation technique in fall detection research.…”