Abstract
Purpose :
We have developed a wearable video camera-based device that provides collision warnings via vibro-tactile wristbands. The device is intended to be used in conjunction with habitual mobility aids. We present results from the first eight participants of a double-masked randomized home-use clinical trial of the device.
Methods :
To date, 1 guide dog user and 7 long-cane users with minimal or no light perception, but with independent mobility, have finished the study. After training with an orientation and mobility expert, they used the device in their daily life for 4 weeks. Unbeknownst to the subject, the device switched between two operating modes in randomized schedules: active (collision warnings provided) and silent (device detected potential collisions but no warnings were provided). While in usage, the device recorded data including videos from the camera, operating mode, and the detected collision warnings. The detected events were reviewed visually by a rater masked to the operating mode, noting the following: true hazard or not, and whether there was a contact with the object – and the nature of the contact (cane or body). The difference in the body contact probability (relative to true hazards) between the active and silent modes was the outcome measure. Wilcoxon signrank test was used for comparison between the two modes.
Results :
A total of 28 hours of video data (median [IQR] 3.5 [3.2] hours/subject) containing 2220 events (194 [258]/subject) were analyzed. The probability of a body contact was significantly lower in the active mode than silent mode (active: 4.9%, silent: 8.9%, p=0.023). The device provided true warnings for 44% of the detected events (active: 41%, silent: 51%, no significant difference between modes, p=0.078). Many of the false warnings were due to body sway when walking along a fence/wall where there was no direct collision risk (but they might nevertheless have helped maintain orientation). About 30% of the true warnings were followed by either a cane or body contact, with no significant difference between the two modes (active: 23% [22%], silent: 31% [15%], p=0.84).
Conclusions :
The reduced probability of body contact in the active mode compared to silent mode is indicative of the real-world mobility benefit of the device. The trial is ongoing and more data is expected to further clarify the findings.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.