More choices on detection analytics

[Paul Green]

Hi, are there any plans to allow us to further hone the analytics to our own situation? Particularly useful for me would be choosing the axis of detection - I'm testing the system with a Hikvision doorbell, and there's a a fair amount of left to right traffic with vehicles on the road outside, but the only thing I'm interested in is people coming towards the door. I can also see a use for people to want to be alerted when someone goes from left to right, but not need an alert when they go right to left, i.e. cars coming, but not going.

[CamectArup]

Direction of movement is something we haven't really thought about, in part because you're the first person to ask. For people who have a camera whose angle can be adjusted (i.e. not a doorbell) I think it usually works to adjust the camera angle a bit and then use positive and negative regions to select the alerts that are interesting. 

Would you be willing to share a shot of what the view through your camera looks like? I think I can imagine scenarios where the region approach won't work well due to the positioning of the doorbell, and I'm wondering if you have one of those. (Have you tried marking off regions manually to see how far you can get with it?) 

[Paul Green]

Hi Arup, See attached. 

I'll read up on masking off the areas and see what happens - I did try within IVMS before the Camect arrived, and due to the camera only being 2 dimensions, tweaking detection zones didn't work owing to the fact the road is right behind where I'd need the face to be detected.

[CamectArup]

Hi Paul, 

     Is your goal to capture only the people walking from the gate to the door, or also to capture people walking towards the gate from the street? If it's the latter, I suspect axis of movement would be hard to use because the street is narrow and most people would approach the gate at an angle, rather than straight towards the camera. 

As far as trying with manually marked-off areas, I think the strategy should be to think about where the feet of people walking up to the door will be. Alert ratings (i.e. including those containing an area) are applied in order from most recent to least recent, and as long as part of an object falls into an area, you have a match. 

So, if the goal is to capture all people walking from the gate to the door, as well as people getting close to the gate, I'd use one area to mark objects uninteresting that covers most of the image, and then add a second area that covers the area in which people, etc should be interesting, and in drawing that focus on where people will step, so that you don't have to include a large chunk of the street.