I'm looking for a method to define level of stress (objectively, i.e. not with questionnaires) at home. 

The idea is to access general stress level of a person once every morning during considerable period of time (e.g. several months) - a non-invasve method instead of blood tests. 

There might be many stressors both psychological and physiological - from financial and family problems to junk food and hangover. The assessment shall be done at home, so simple means are required. It's possible to use home devices like home HRV and home EEG but both don't measure stress reliably (HRV does not measure mental stress at all, and home EEG are not reliable enough).

I thought about computer cognitive tests - but If a person performs the same test day after day then his results may improve not because of stress decrease but because he has improved his playing skill. And so the results can't tell if his stress level goes down or up.

Any ideas?

 

Might be worth looking at handwriting analysis.

I'm looking for an automatic software soultion. Does it exist in the area of handwriting analysis?

You could look into the technology used in "lie detectors" - they are utterly useless for detecting lies, but they may have some correlation with stress more generally

Heart rate

Blood pressure

Galvanic skin response

Cortisol test (using saliva and/or urine)

Breath (plasma adrenaline can be measured using HPLCy with fluorometric detection)

 

A team at Stanford also created a dermal patch that detects cortisol levels: https://www.mobihealthnews.com/content/stanford-researchers-create-cortisol-detecting-wearable-measure-stress-health

Caltech took a different approach to measuring same: https://www.medgadget.com/2020/02/first-noninvasive-cortisol-detector-to-monitor-stress-mental-conditions.html

There are different types of stress that correlate with different measures. There is also a difference between physiological stress responses (such as the mentioned cortisol levels) and the (mental) perception of stress. In other words, one has to define rather narrowly what one wants to figure out as something like universal stress does not exist as such. 

For example, there are wearable devices out there that indicate stress using ECGs and EMGs, but typically they are only somewhat predictive of immediate stressful situations (i.e. time-limited fluctuations above a baseline). Cortisol can also be indicative of long-term (i.e. chronic) stress, however the baseline levels need to be established for each individual and can be quite tricky. Many indicators undergo changes during the day or are heavily influenced by other physiological changes making data-driven diagnoses difficult. 

There are plenty of studies including on wearables that OP might want to look into. Much of the recent focus has been on the data analysis side rather than the measurement side. I.e. can we use the wealth of collected physiological data (e.g. from wearables or other sources) to accurately predict the physiological state of the individual? While there are folks selling it quite successfully, my reading is much more skeptical on that matter for now.