How would I run a pre-trained object detection TensorFlow model on an ESP-32 Cam

I am currently working on a project that will see my dog in a specific area in my house, it will beep for me. I made a prototype on a raspberry pi 4 and it worked great, but now I’m having trouble shrinking it down. I know an ESP-32 Cam is my best option since it has Bluetooth, WiFi, a camera, and is really cheap, but I’ve looked everywhere on how to do this, but am coming up with nothing. I know what I’m trying to do is perfectly possible, I just don’t think its been done before(well at least from what I’ve found from my countless google searches). I need TensorFlow to run independently on the ESP-32 without the need to be connected to a web browser(so no TensorFlow js that runs from the local IP address, unless if there’s a way to do it without having a separate device connected) I want to be able to just plug it into an outlet and it will start working. If you have any questions or confusion, don’t hesitate to ask. I need all the help I can get. Thanks.

Hi, I was trying to so something similar but face detection on ESP-32 CAM. I tried to research for how to such a thing without to be connected to a web browser. So, I was asking if you find anything useful about this?

I was thinking to download the TensorFlow.js library, which is 1.3 MB, and put it in the PSRAM of ESP32 and any script that needed to be call. So, this theoretically should be a problem to let the HTML file to get the script from the PSRAM, but I didn’t try it yet.