Devlog: Visora is Finally Done !!!

After roughly 40 Hours spread across the last 7 days, Visora is finally working.
I genuinely thought this would be one of the easier parts of Gesture Link.
I was very very wrong.

Visora is the vision module of the project. The goal is simple: use an ESP32-CAM, run object detection using YOLOv8, figure out what’s around the user, and then provide audio feedback.

Sounds straightforward until you actually try building it.

The biggest problem was that nothing wanted to work consistently.
One minute OpenCV would connect to the camera stream perfectly:

cap = cv2.VideoCapture(CAM_URL)

The next minute it would just sit there refusing to open the stream.

while not cap.isOpened():
     print("Cannot open stream...")

I spent hours thinking my code was broken when it turned out to be network issues, camera settings, and random ESP32-CAM problems fighting each other.

Then came YOLO.

Getting the model itself running wasn’t too bad.

Getting it to run smoothly with a live camera stream was another story.

results = model(frame, device=device)[0]

Sometimes detections worked.

Sometimes they didn’t.

Sometimes the frame processing would slow down enough to make the whole thing feel unusable.

My favorite bug was when everything looked correct. lol

• The camera worked.

• The AI worked.

• The ESP32 worked.

• The code ran without errors.

And yet nothing happened.

After staring at the screen for far longer than I’d like to admit, the issue ended up being something completely unrelated to where I was looking.

What makes this extra painful is that the final code looks so simple.

You look at it and think: That’s it?

But behind every line is probably an hour of me wondering why it wasn’t working.

With Glove A converting sign language into speech, Glove B converting speech into Braille through vibration motors, and Visora finally working, Gesture Link is now fully complete.

7 days
40 hours
Way too much debugging
** Gesture Link is Officially Done !!**