Fall detection is a typical application of both medical expert systems and wearable expert systems. There are four primary types of falls, i.e., left-sided horizontal, right-sided horizontal, supine, and prone, whereas most of the previous studies and commercial systems focus on distinguishing fall or not.  

Accurate and rapid multi-directional fall detection is of great significance for providing the human body with timely inflatable airbag protection in the fall direction and helping doctors shorten the reaction time to assess the severity of particular joints. 

In a study published in Expert Systems With Applications, the research group led by Prof. DAI Houde from Fujian Institute of Research on the Structure of Matter of the Chinese Academy of Sciences reported a novel fall detection methodology based on smart insoles and a long short-term memory (LSTM) framework with a trained referencing denoising auto-encoder (RDAE).  

The automatic feature extraction via RDAE, instead of manual offline ex-traction, has the advantage of without subjective labeling and analysis, thereby more likely to select near-optimal features and enhance the detection robustness. 

The researchers employed a pair of wireless in-shoe insoles of which each side is equipped with 13 plantar pressure sensors and a tri-axial accelerometer, to capture comprehensive spatial-temporal gait parameters. Hence an effective response to a fall, together with the estimation of the corresponding direction, can be accomplished, where the accuracy and response time are two primary concerns.  

The proposed RDAE-LSTM network provides a reliable testing result in classification, with 98.6% accuracy and 8.7 ms response time for determining fall directions, demonstrating a more convincing performance than that of other algorithms.  

This methodology is an unobtrusive choice for users whose daily life is not affected by the fall detection device. The RDAE-LSTM model was proven to accurately and quickly recognize falls in four directions for the unbalanced fall detection dataset. 

This study has great prospects in Internet-of-things applications in the aging society.