Hi WildLabs, hi All,

It's been a while since I've posted on WildLabs so please be kind. 

I'm wondering if there is anyone in this space developing wildlife AI models to be used on Full Motion Video? 

I've got a video camera hardware solution that can process 30 frames per second, but so far have not found an AI model that would run on it in real-time for detection, recognition, classification and tracking of wildlife. 

Here's the pitch: since moving on from ZSL and the Instant Detect 2.0 real-time camera trap system, I've been working at a company called Overview that builds an AI-enabled pan/tilt/zoom sensor system called PodView. It’s essentially a powerful edge-computer with a motion platform and sensors mounted on top. 

In the base of PodView is a fully integrated NVIDIA Jetson Orin AGX processor, which is connected by fibre-optic sliprings to an Electro-Optical 50x zoom visual video camera and a Long Wave Infra-Red (LWIR) Thermal Imager in the head. 

The motion platform is extremely agile and accurate, so it can point at any two points in 3D space within 0.5 seconds with a >0.01 degree accuracy; while the glass-to-shared memory time is ~90ms with the full control loop ~115ms. Combining this extreme low latency with this level of agility means that with the right AI model you can automatically detect and track really fast moving things.

PodView was originally spec'd for Defence but is actually a dual-use product. As such, it's a very robust and rugged platform and has a very low maintenance requirement. It runs off 48V DC so can be battery/solar powered, and connects through IP so can use any communications bearer (4G, Starlink, etc). 

It also has a user interface to control the camera, set up automatic Scans and Tours of an area and can also set-up different ‘Rules’ for any installed AI, e.g. Scan this area and if you see X, report to Y and track it.  

I think there is a strong use case for PodView to be deployed on tall infrastructure within protected areas where it could provide the ability to remotely or autonomously look in all directions, zoom in and out, and use AI powered recognition and classification to only generate alerts when things of note are seen. 

My reason for posting here is that Overview don’t build AI for PodView. We say it's AI agnostic. Typically our customers install their own AI payloads and we are slowly building up a number of AI payload options developed and maintained by partners. 

So far I've not come across any wildlife focused AI models that would work with full motion video, so would be really interested to hear from anyone if they exist. 

Best wishes,

Sam