Sub-Project A:  Video-based recording of real-life falls

Research team: Greg Mori (co-leader), Fabio Feldman (co-leader), Shahram Payandeh, Joanie Sims-Gould, Stephen Robinovitch

The goal of this project is to determine “how” and “why” falls occur in older adults, based on careful review of video footage of real-life falls experienced by residents of two partnering long-term care facilities. Video footage is acquired from a network of over 400 digital video cameras installed in common areas (dining rooms, lounges, and corridors) of two partnering long-term care (LTC) facilities in the Vancouver area (Delta View and New Vista Society care homes). In the event of a fall, an incident report is completed by care providers. We then review these reports to identify the location of falls, and retrieve the corresponding video footage. 

 

Figure 2. Snapshots from video capture of real-life falls in seniors, acquired under the TIPS research program. These videos show examples of successful (top) and unsuccessful attempts to arrest a forward fall with the upper limbs.
 

We then carefully analyze the video footage to determine the cause and circumstances of each falls, based on a structured questionnaire completed by an expert team. Characteristics of interest include (a) the cause of the fall (e.g., slip, trip, syncope, or loss-of-balance), (b) the activity at the time of the fall (e.g., walking, turning, reaching, or standing), (c) the direction of the fall, (d) the use of specific balance recovery or “safe landing” responses (e.g., stepping or grasping, or arresting the fall with the outstretched hands), (e) the presence of clutter or apparent role of environmental factors (e.g., tripping hazards, poor lighting), and (f) the type of assistive device being used (if any). In related work, we are developing computer-based methods to automatically identify falls (and distinguish key characteristics of falls) from video data. Our long-term goal here is to develop video surveillance systems suitable for extensive implementation in high-risk environments, which provide automatic detection and alerting to care staff of fall events. These efforts are based on novel algorithms that recognize human actions based on “novelty motion estimates.” 

Preliminary results from this Sub-Project challenge some common perceptions concerning the circumstances of fall in older adults. For example, we are finding that most falls are caused by incorrect weight shifting (or transfer of body weight), which is more than twice as common as the next most-frequent causes (trips, hit/ bump, and loss of support with an external object). We have also observed that head impact occurs in over one-third of falls, and that while the arms are commonly used in an attempt to arrest the fall, this response is insufficient for preventing impact to the head, perhaps due to insufficient upper extremity strength. Ongoing analysis is providing a unique window into the mechanisms of fall initiation, descent, and impact.