Motion analysis for in-home gait and balance assessment using inexpensive video sensors

Abstract

Gait has been studied in recent years as one of the newest topics in computer vision as it offers potential for applications in various areas, including biometric identification, medical diagnostics and therapeutic tools. Clinical research has identified a clear link between human gait characteristics and different medical conditions. We are particularly interested in gait and balance analysis for fall risk assessment in elderly people. This dissertation proposes the use of multiple inexpensive web cameras for three-dimensional gait and balance analysis in a voxelized space. The main advantages of this approach include low cost, unobtrusive monitoring compared to wearable sensors, and three-dimensional voxel reconstruction which provides a non-camera view-dependent capability. It is thus suitable for daily assessment and continuous monitoring of elderly people's non-controlled daily living environment. The gait parameters studied include walking speed, step time and step length; balance was assessed through body sway. In addition, 180° turnaround time and number of steps are also studied. These parameters were obtained from a three-dimensional voxel reconstruction, which is built from two calibrated camera views. These parameters are validated with a GAITRite Electronic mat, a Vicon motion capture system in the lab, and a physical therapy expert rating. Excellent agreements were found for walking speed, step time, and step length between the webcam system and the GAITRite, with no statistically significant difference. There is no significant difference in walking speed and step length comparing the webcam system with the Vicon. Significant difference was observed for step time between the webcam system and the Vicon, as well as between the GAITRite and the Vicon. There is no significant difference in walking speed between the system and the expert, but highly significant difference in step time and step length, caused by the expert's strategy of rounding up the number of steps when there is an incomplete gait cycle. Overall, the walking speed was found to be the most robust and reliable gait parameter captured through all the tests. The body sway and turnaround parameters were also found to have a reasonably good agreement given the limitation of frame rate and voxel resolution. The system was further tested and validated with assistance of elderly residents from TigerPlace, an independent living apartment complex for seniors in Columbia, Missouri, in scripted scenarios representing everyday activities. The residents were first tested on a GAITRite mat with cameras recording images. The extracted gait parameters from the camera system were compared with the GAITRite; and excellent agreement was achieved. The residents then participated in the scenarios, with only the cameras recording. Parts of the scenario video of 10 participants were evaluated and rated by a physical expert in the research team. The algorithm results matches well with the expert rating. Based on these good validation results obtained in a realistic non-lab setting, further analysis is therefore reliable. Results on 18 elderly participants were analyzed. The analysis showed that the residents displayed different gait patterns when the realistic scenarios were compared to the GAITRite runs. Furthermore, the complete set gait parameters provided a clear picture of a person's gait status and were used to screen out people at risk of falling. This research shows how long-term gait monitoring can provide trend information on a person's gait characteristics and proves the importance of continuous gait assessment in a daily living environment. Research results indicate that the development of this technology can provide capability of monitoring gait and balance in daily living environment for elderly people, which in turn can be used as part of a continuous balance, stability and fall risk assessment tool.