ESP32 PSRAM Wifi Configuration C++

[richard_gotthard123]

Question/Issue:
I’ve used this repo for deployment towards ESP32 PSRAM Timer Camera (OV3660).

I’ve managed to flash and use the serial protocol for “run impulse” and run daemon towards our project, but for the next stage i want to deploy the device and send the stream towards our server. Do i need to set up a Wifi configuration outside of the firmware to best implement this or could i implement something inside the AT serial protocol to best solve this?

Project ID:
183850

Context/Use case:
Integrate wifi configuration for deployment with ESP32 PSRAM Timercam. Set up Wifi-configuration with our server.

[shawn_edgeimpulse]

Hi @richard_gotthard123,

The firmware you linked to is meant to run standalone on the ESP32 where you can communicate via AT commands. You have a few options:

• Use another microcontroller to talk to the ESP32 via AT commands. However, note that the ESP32 does not have WiFi running in that standalone firmware, so your co-processor would need to have its own WiFi.
• Modify the standalone firmware. There’s a lot in there, so this might be quite an undertaking.
• Start with a simpler example for performing inference (e.g. GitHub - edgeimpulse/example-standalone-inferencing-espressif-esp32: Builds and runs an exported impulse locally (ESP IDF) or GitHub - edgeimpulse/example-esp32-cam: Builds and runs an exported image classification impulse on ESP32 Cam) and write your WiFi application around that.

If it was my project, I would go for the third option so that you can run inference and networking on the same ESP32.

[richard_gotthard123]

Hi!

Thank you for reaching out about a solution! I figured i would use the third alternative like you described. How would i be able to “adding sight sensors” to its feature array if i would be interested in that, in other words sending my bounding boxes or images to it for processing?

[shawn_edgeimpulse]

Hi @richard_gotthard123,

In the second link I posted, if you go into the “Basic Image Classification,” you can find an example of a function that captures an image from the ESP32 camera, resizes it to 96x96, and converts it to an RGB888 format. That function is here. Note that it also contains HTTP code for streaming the image (you may or may not want that).

In the main loop, the function run_classifier() is called to perform inference. You need to provide a callback to that function that feeds data from the image buffer. An example of that callback is here. This reads raw bytes from the resized_matrix (which stores the resized image) into the signal_ptr buffer, which is used during the inference process.

After you call run_classifier(), your results will be stored in an ei_impulse_result_bounding_box_t struct, which you can read about here. You can use the label, x, y, w, h, and value from that struct to determine where the object(s) are in your image.

Hope that helps!

[richard_gotthard123]

Sorry for reaching out late but thank you again for the feedback! Our team got stuck inside another problem.

We wanted to implement the solution inside our ESP-IDF project similarly to how this is implemented due to it being easily compatible with C++ library from the Edge Impulse Studio and us comparing it the available firmware firmware. Would using arduino still be advised as a solution for running the classifer?