Context- and Terrain-aware Gait Analysis

Overview

The average lifespan of individuals in many developed countries is increasing. This factor paired with the increase in global population has the potential to put a strain on healthcare systems with regards to age-related conditions. Particularly, this research considers the impact that conditions such as Parkinson's disease, dementia and stroke have on the walking capabilities on affected individuals. This research project aims to obtain a gait analysis dataset consisting of sensor data captured during regular daily activities on common terrains such as grass, paving slabs, gravel, etc. The dataset will be collected with a custom sensor system which captures mobility data from a cohort of healthy controls of all ages and people with dementia, Parkinson's disease, stroke survivors, multiple sclerosis, etc. Various machine learning algorithms (custom-implemented using Python) will then be used to determine the walking activity (walking, ramp ascend/descend, stair ascend/descend etc.), the terrain (grass, pavement, carpet etc.), and various walking-related parameters (step length, step height, cadence etc.). It is our hope that these features will enable remote gait analysis to be performed with sufficient contextual information to enable remote diagnosis and rehabilitation tracking for those at risk of falling.

Walking is a crucial ability to allow people to live normal, healthy lives. However, various conditions which affect the brain or the body such as Parkinson's disease, dementia, stroke, multiple sclerosis, and amputations threaten a person's ability to walk and can lead to falls or the fear of falling. Either of these fall-related burdens can severely affect the quality of life for a person, particularly those who are most vulnerable, such as older people. To detect fall-related issues in a person's manner of walking (their gait), a process called gait-analysis can be performed which involves a team of specialists using video cameras to record someone walking in a laboratory environment and analyse the video to identify problems. However, current technology is rapidly advancing towards the capacity for remote gait analysis, which uses wearable sensor technologies to capture one's gait. This provides many benefits such as a more natural walking style, automatic data analysis, and reduced time needed by specialists to perform the analysis. The largest of these benefits, however, is the capacity to wear the device outside of the laboratory to see how a person walks on real terrains. Many current studies have shown great strides in producing highly accurate gait analysis systems. However, a real-environment dataset for these systems to be tested on does not yet exist. Furthermore, datasets including a range of people with conditions that increase their risk of falling are scarce and typically only focus on one group. This study aims to produce and analyse the first real-world gait analysis dataset which includes a wide range of gait-affecting conditions, and to highlight what worked and what didn't for future researchers to build off when designing and implementing practical solutions to real-environment gait analysis.