Fall detection lags fall by quite a bit

[truedat101]

I am comparing fall detection options, and a sensor designed to detect falls seems like a nice approach. The alternatives:

• human monitoring (i.e. live human watching that other human doesn't fall) . Latency is probably milliseconds (<= 300ms) : https://humanbenchmark.com/tests/reactiontime
• ML based computer vision algo using >= YOLO8 , should be near realtime , maybe within 1sec
• IMU based detection using a "falling" motion detection : maybe < 500ms ... many ways to do this, but requires the targets to wear an IMU, may have false positives, calibration and orientations need to be known
• MMWave Sensor fall detection: with default parameters, it seems to need about 10 seconds to actually log a fall. And it seems like I need to flail and put my feet in the air to make trigger. The sensitivity is set to 15.

Can you comment on the tuning of the approach so that the sensor is on par performance wise? I prefer something like an mmwave sensor from a cost/compute perspective. Cameras have aspects like lighting, FOV, focus, privacy, and compute cost which make it slightly less appealing for deployment.

Code used to test is here : https://wiki.seeedstudio.com/getting_started_with_mr60fda2_mmwave_kit/#fall-module