Step counter is activated just by hand movements

Those are both normal things with garmin watches when worn on the wrist.  Steps are based on what the watch sees.  For the second one, this happens when you are pushing a shopping cart, stroller, or using hiking poles too,  

The work around in both cases is to put the watch in your pocket when you are in those situations.

Another thing, what activity are you recording exercises on the floor?  Non step based activities don't count steps on some devices,  

jim_m_58 said:

The work around in both cases is to put the watch in your pocket when you are in those situations.

... or wear the watch on your ankle.

Avoima , _psx_

But seriously, how is a watch worn on the wrist supposed to accurately detect leg movements? Surely that can only be possible indirectly if usual physiological acceleration patterns are detected.

What activitry are you using when exercising on the floor?  

Yes, in order to accurately count steps (and not count steps when no steps are taken), the acceleration patterns would have to be analyzed. In my view, it is almost certain that the acceleration patterns for 'hand movements + steps', vs. 'hand movements only', vs. 'steps only' are different - even when the measurements are taken by a wrist-worn watch.

The only question is, how to detect those differences and to correctly and reliably categorize these three types of patterns. 

Avoima said:

The only question is, how to detect those differences and to correctly and reliably categorize these three types of patterns. 

I accept that without wearing a sensor on the shoe or ankle, it's simply impossible to reliably detect steps in all situations.

The human body isn't rigid and there are almost independently 5 joints between the watch on the wrist and the sole of the foot. And the step pattern of each person may be individual and also depends on the cushioning of shoes and many other things, for example, walking with lowered arms, talking or looking on the phone while walking, walking up or down a steep path, floors, at home, doing squats, jumping on a trampoline, ...

So it is only possible to implement a general acceleration algorithm in a small clock with limited execution power and memory.

Cardio and no activity there is no differences with steps. But of course I move my body and hands.

Workaround doesn't work because I think watch check if it's  on wrist, If not (there is no HR e.g.) no step counting.

Well, unless you have an external HRM, there will be no HR, and isn't "no steps" what you want?

I've only written that probably there is no step while you have watch in pocket because watch check if is on wrist so your

jim_m_58 said:

The work around in both cases is to put the watch in your pocket when you are in those situations.

won't run. But I've observed watch during pushing not shopping cart but  lawn mower :) and it count steps, so it's good.