Results 1 to 7 of 7

Thread: Cameras for motion detection inputs to presence sensing

  1. #1
    Automated Home Legend chris_j_hunter's Avatar
    Join Date
    Dec 2007
    Location
    North Lancashire
    Posts
    1,713

    Default Cameras for motion detection inputs to presence sensing

    we're now running nine cameras in our set-up - would be more, but hardware problems temporarily keeping two more out ...

    meaning we've tried including motion detection in presence sensing ...

    mixed results so far, partly because lights going off can be seen as motion !

    so, wondered, anyone else trying this - any successes ?

    Chris

  2. #2
    Automated Home Guru Nad's Avatar
    Join Date
    Jul 2008
    Posts
    145

    Default

    I've had a go and I have to say that I'm not having much joy at all

    To put things into a bit more context ...I have 2 cameras at the front of the house, one looking over the drive way and one over the door. The one that I was most keen to use motion detection with was the one over the drive way so I can use it to control the front lights. All good in theory but in practice the motion detection is (in my case) horrendously inaccurate and I have all but given up on it for now.

    Below are the main reasons i have noticed for false readings on my driveway camera...

    1. Shadows - This is a huge issue for me as the orientation of the drive way means that people walking past tend to generate quite pronounced shadows that go across the front during most times of the year.
    2. Reflections - Things like other cars, people and clouds reflected off the glass on our cars.
    3. Lights - We are not too far from a junction so cars turning at night will cause their headlights to go across portions of the driveway.

    I'm no expert in the field of CCTV and motion detection algorithms but I guess it may be possible to reduce the effect of some of these by allowing the sensitivity of the motion detection to change depending on time of day, light levels and position of sun but I think the most effective solution would be to have the frames be processed by some form of AI so it can be taught the differences. It may be possible to then take it up a notch and teach it who certain people are and various scenarios like someone trying to deliver something and what the delivery company could have been (e.g. Royal Mail, DPD, TNT, etc.).

    Thanks,
    Nad

  3. #3
    Automated Home Legend chris_j_hunter's Avatar
    Join Date
    Dec 2007
    Location
    North Lancashire
    Posts
    1,713

    Default

    yes, our one that looks out across our front yard has some of those issues, too ...

    the grid overlay approach to motion-sensing works really well, for excluding (eg) people going-by from being detected - but, at times of the day when the Sun is low & to the east / south-east, shadows from passing cars & people can be thrown into other areas of the image, triggering motion detection ...

    ditto when people opposite are manoeuvring their car at night, its lights can do something similar ...

    and when our yard lights go out, they can be seen as motion, too (presumably off is hard-edged, whilst on is sufficiently soft-edged to be missed) ...

    indoors, too, we find lights going off can do the same, though we might have a chance of creating some General Logic to discount the effect, because Cortex generally knows what its Lights are doing ...

    as you say, it's tempting to believe some pattern recognition coding could help - neural net or otherwise - 'though have no feel for what would be the necessary resolution in the images to achieve this, plus using WiFi to communicate with cameras brings issues affecting images, which could complicate things ...

    being able to allocate grid elements to named zones, for dealing-with separately, would be more than useful - ie: motion in one area will often need a different response to motion in another (perhaps depending on time of day & circumstances, too) ... plus it might provide a basis for simple pattern recognition to deal with (eg) the shadow & car headlights issues ...

    Chris

    PS: another false trigger we get, with a camera looking out, is raindrops running down the window, when it's raining a lot / heavily ...
    Last edited by chris_j_hunter; 6th January 2017 at 02:47 PM.

  4. #4
    Automated Home Guru Nad's Avatar
    Join Date
    Jul 2008
    Posts
    145

    Default

    This got me thinking and I was pondering about the possibility of off-loading the motion detection part to another application and then feeding that back into Cortex so off to Google I went ... as you do ... and came across this:

    https://www.sighthound.com/products/sighthound-video

    Looks quite interesting so I may give the trial version a go and see how that works out and to see if there are any options to pass detected motion to Cortex. I've seen that there is a plug-in for Homeseer so it is more than possible. The other two options (Sentry and Cloud) look quite interesting too.

    Thanks,
    Nad

  5. #5
    Automated Home Legend chris_j_hunter's Avatar
    Join Date
    Dec 2007
    Location
    North Lancashire
    Posts
    1,713

    Default

    no feel for it, but imagine it might need relatively hi-res cameras, to work reliably ...

    our Mobitix & Y-cam might be accommodated (the brands are on their list), but we're mainly using tablet cameras (some front, some rear) ...

    will be v.interesting to hear how you get on !

    Chris

  6. #6
    Automated Home Legend chris_j_hunter's Avatar
    Join Date
    Dec 2007
    Location
    North Lancashire
    Posts
    1,713

    Default

    just to add an extra factor ...

    ie: with the mist / fog we have today, that is not too terrible, can still see things a few hundred metres away - we're finding motion sensing is much less effective at noticing things moving outside ...
    Last edited by chris_j_hunter; 7th January 2017 at 05:33 PM.

  7. #7
    Automated Home Guru Nad's Avatar
    Join Date
    Jul 2008
    Posts
    145

    Default

    All in all sighthound has been pretty good. The detection algorithms used are very good indeed and it does do a very good job of identifying people (the camera angle plays a big part in accuracy and detection ability, kind of obvious when you think about it). The ability to set a size of object to ignore is also very handy as it means things like leaves blowing or small animals in the drive won't trigger an event. I wonder if Cortex can have that kind of feature added

    Unfortunately it's still not as good as I would like. The main gripe I have is that you can ask it to look for people, other objects or anything else. Which is fine unless you want to know the difference between a person and a car for example. I managed to get around some of it by using multiple rules but that did increased the false detection rate. I may have been able to improve on it a bit more but the trail license ran out.

    The other thing that is against it is the cost. The basic version only supports 2 cameras and is $60 which is not too bad but then you jump to $250 for the Pro version which gives the ability to use unlimited cameras. I can't justify $250 on this software if I have to add other sensors to reduce the false detection rate but $60 for 2 is almost acceptable. If it was $60 for 4 cameras, then I think I would probably have gone for it.

    So in conclusion, Sighthound is actually pretty good and does a good job of detecting people. The CLI output also lends its self quite well to integration with other systems (in theory as I didn't play with this). However, I'm going to park this for now and I'll probably revisit this topic in the not too distant future to see if things have moved along but for now I'll be sticking with Cortex for the HA side of things and either the HIKVISION NVMS or some other NVMS for the video side of things. The HIKVISION offering is a bit too coarse when it comes to motion detection but storage is getting cheaper and cheaper so I just let it record 24/7 and then set it to log event triggers so it makes it easier to look for things.

    How I think I could make a workable system using Sighthound:
    1. Have multiple cameras.
    2. Have multiple rules.
    3. Have physical sensors.
    4. Feed it all into Cortex to do the action derivation based on all the info from above.

    I think that's enough rambling

    Thanks,
    Nad

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •