Wildlife Trap

I would like to create a wildlife trap system.So we will start with OpenMV Cam H7. I am planning to add a PIR Sensor, IR illuminator, LTE comm interface. Basically i am trying to make it up for 1-2 months (including sending of the still images)
As i want to make it more power efficient to start the board with a PIR sensor, so that we will have a longer standby(may be power down). (although Standby consumption - OpenMV Products - OpenMV Forums it says the consumption in stand by is 1.6mA is not bad.)
I need advise on the design principles and integration of the system. I have evaluated the products like http://www.willfine.com/index.php/products/40cg. But it does not allow to customize the system and software for different analysis.

If you have the budget I’d reconmend using the FLIR Lepton 3.5 and our FLIR Lepton adapter. Thermal vision will work during the day and night and it will also make it trivial to see the animal in the dark.

If you can’t use a FLIR Lepton then differential frame differencing probably + TensorFlow image classification is what i’d do.

Basically, if you subtract the current frame minus the previous frame that was taken like a second ago you get a movement image that you can then Train a CNN with that can detect animal blobs in the image versus random noise. To create the data set you’d need to deploy the camera in the wild to get these difference images.

Alternatively, you can run a CNN on the raw image… however, you’ll need to make the CNN work on night and day images with things like glare in the background and etc.

Anyway, the problem is easy with the FLIR, harder with a regular image sensor.

Hi there, any updates on this thread? Did you manage to use the thermal board to detect wildlife? Probably it will only work on “warm-blooded” animals. Would it be possible to extend the H7 plus board with a second connector to plug two sensors and switch to a the RGB camera after detection with the FLIR sensor?