monitoring presence
  • I'm using SS5 to detect vehicles in a rural driveway, with great success, thanks (using the Indigo plugin). But without manually inspecting a camera image, it isn't easy to infer if a car has arrived, left, or just moved.

    I'd really like to monitor whether vehicles are present or not, by periodically triggering the SS recognition algorithm to fire.

    Is there a way to achieve this through the API?
  • Great to hear the vehicle detector is working well for you.

    Unfortunately there is no way to do what you are describing, and it would not be simple to implement under SecuritySpy's current scheme. The way it works currently is that motion detection is performed first, which identifies the moving object - which typically is just a small part of the whole video frame. The image is then cropped to just the moving object, which is fed to the AI for classification.

    So, you can't just send the whole frame to the AI, as this is not what it expects. Objects would be too small (as a proportion of the whole image) to identify accurately.

    Of course, there are standard approaches in computer vision to do what you are describing (e.g. a moving window across the whole frame, or a first-line neural net to identify bounding boxes of objects prior to classification) but this is not what SecuritySpy does at present.
  • This would be a cool feature. so it basically keeps count of number of cars (detected objects) in the driveway.

    How intensive would the classification be if it was done constantly rather than just on motion?
  • Classification is a CPU-intensive operation: classifying one image typically uses most of the CPU's resources for around 100-300ms (depending on the speed of the Mac, and whether any of this can be offloaded onto the GPU - something that is becoming more common with newer Mac models). So it's something that must be done as infrequently as possible, and certainly can't be applied continuously to a video stream in a multi-camera environment.

    If the classifier were to be coupled with a static object detector, this could be run frequently (say once every few seconds) to maintain a count of interesting objects, and this would probably work well, but it's not something that SecuritySpy can do at the moment, as it lacks this static object detector. SecuritySpy's current detection capabilities are based around moving objects only, as this is generally going to be the most important thing for most users.
  • thanks for clarifying the current limitations.
    in my use case, I'd be happy to trigger the routine infrequently...say every several minutes (so compute resource is probably not such an issue). An example would be check for a vehicle presence before shutting a gate, turning off lights at a particular time if no vehicles are present.
  • thanks Ben

    Object detection would be a nice feature. I know it could be done via say the Google Vision API but costs would be hight if done frequently.

    Looking at the motion capture options, I wonder if a "pre-capture" option can be added image capture, that way you could take the first and last capture images and compare outputs.

    To be fair, maybe post-capture on its own might work is well, maybe have a less frequent object detection update and then have another check done after motion ends.

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!