How to Use YOLOv8 with Live ESP32-CAM Stream for Eye Detection?

[sipov]

Hi, I’m trying to use YOLOv8 with an ESP32-CAM. I want to stream video from the ESP32-CAM to a local server, and then run object detection using YOLOv8 in Python.

I wrote the following code:

from ultralytics import YOLO

url = 'http://192.XXX.XXX.XX/capture'

model = YOLO(r'...\weights\best.pt')

results = model.predict(source=url,
                        show=True,
                        show_conf=False,
                        conf=0.2,
                        imgsz=320)

However, the code doesn’t work as expected. I'm not very experienced in this area, so I’m not sure what I’m doing wrong. I assumed that the predict() function could take a video stream URL directly.

My goal is to:

1. Send video from the ESP32-CAM to a server on the network.
2. Use Python with YOLOv8 on that server to process the stream and perform real-time prediction.

Am I trying to do too much considering the hardware limitations of the ESP32-CAM?
Should I approach this differently?
Any help or guidance would be greatly appreciated.

Thank you!

[glenn-jocher]

Your ESP32-CAM URL https://www.quora.com/I-have-an-IP-cam-and-its-connected-to-192-168-0-xx-but-Im-mostly-connected-to-192-168-1-xx-Can-I-access-it-through-main-one appears to be a single image capture endpoint rather than a video stream. YOLO's predict function expects either a streaming URL (like RTSP/MJPEG) or requires you to handle the image capture loop manually.

For ESP32-CAM streaming, try using the stream endpoint (usually /stream) if your firmware supports it, or create a loop to continuously fetch from your capture endpoint:

import cv2
from ultralytics import YOLO

model = YOLO('best.pt')
url = 'https://stackoverflow.com/questions/37158506/errorconnection-to-http-192-xxx-xxx-xx8080-refused:81/stream'  # Try stream endpoint

# For MJPEG stream
cap = cv2.VideoCapture(url)
while cap.isOpened():
    success, frame = cap.read()
    if success:
        results = model(frame, conf=0.2, imgsz=320)
        annotated_frame = results[0].plot()
        cv2.imshow("ESP32-CAM", annotated_frame)
        if cv2.waitKey(1) & 0xFF == ord("q"):
            break
cap.release()
cv2.destroyAllWindows()

Check your ESP32-CAM firmware documentation for the correct streaming endpoint URL and ensure it supports continuous video streaming rather than just single image capture.