We make edge AI cameras. After focusing on services for a while, it has been a different type of journey to switch to the product. We did an open source pilot product on esp32, and got surprisingly more interest than we thought, so now we are working on a high performance (4k, 60fps, AI chip) device.
Luxonis is way more powerful and has an Intel VPU (Movidius). It's not really meant to be a standalone platform, so it needs a host board (Linux SBC like the Pi). It takes a lot more power, but it can do 60fps on small Yolo models. Its resolution is a lot higher as well.
ESP32-S3 has a pretty small memory capacity, and doesn't have H264-H265 hardware encoding, so you'll be on low-res, low-fps. It only does MJPEG (AFAIK) streaming, so you'll also have to deal with high latency if you want to send that data somewhere. The big bonus is that it's low-cost and low-power, and you're running it directly on the core without an OS.
This means you can do stuff like sleep the cores until something wakes it up (like a PIR sensor that detects people), and it will start streaming in a second or two.
TL;DR: Luxonis stronk, but needs big batteries or plugged in. ESP32-S3 can run on small batteries or solar.
By "AI" here I mean cameras that support running machine vision models directly on the device. We provide the device and firmware, and then on the application and model level it is up to the user.
Lite-esp32 camera https://www.crowdsupply.com/maxlab/tokay-lite Source-code: https://github.com/maxlab-io/tokay-lite-pcb
Pro camera updats will be posted here: https://maxlab.io/store/tokay-riscv-camera/