NightHawk Fall Detection
Results from 3 saved test videos: the original video, our first attempt, and the final version.
Falls caught
2 of 3→3 of 3
Frames showing the alert after the fall is detected
21 of 156 (13%)→156 of 156 (100%)
Practice test on 70 public videos: falls caught
28 of 29→30 of 30
False alarms on 40 normal videos
Not tested→0 of 40
1. Original video
The test videos as recorded.
2. First attempt
The system lost people during fast movement and when they were lying down. First-attempt videos start with an intro screen.
3. Final version
The same videos after the fix. Once a fall is detected, the alert stays on until the person gets up.
What we changed
- The system keeps following a person when it is less sure, such as when they are blurred or lying down.
- It can find the same person again after losing them for a moment.
- Once a fall is detected, the alert stays on until the person gets back up.
- The video now shows only the person, FALL DETECTED and the time.
What this result means
- We made the fix using these same 3 videos, so 3 of 3 is not yet proof of real-world accuracy.
- The 40 normal videos were also used to teach the system. The next step is to test on new videos.
- The tests used saved videos only. The system does not run on live cameras yet.
- This is a proof of concept, not a certified safety device.