Capture images with Grove Vision Ai sensor

1. What is it?

This is a camera-based sensor that supports in-camera Yolo V5 ML models. So that your data never leaves the camera. And you can create them yourself

2. Why build a dataset?

You can find many datasets online. Like from universe.roboflow.com.

But if you have a particular problem it is good to build it yourself. Using the work application camera. Because

• different cameras take different pictures (angles etc.)
• light conditions can be very different
• special subject, not much public data

3. How?

3.0 If you use Linux

Linux (Ubuntu etc.) may not recognize it as a USB device. Therefore we need to create new rules file. File name: 60-visionai.rules

File content:

SUBSYSTEM=="usb", ATTRS{idVendor}=="2886", ATTRS{idProduct}=="8060", MODE="0666"

Save the new rules file to the /etc/udev/rules.d/ directory.

Restart the udev rules:

sudo udevadm control --reload-rules

3.1 Firmware

There is a special firmware file that can be used to capture images. You can download it from here. File name: capture_images_Grove_AI_firmware.uf2

Connect the Grove Vision Ai module with a USB cable. Double click Boot button. GROVEAI will be mounted as a disk drive. Drag and drop firmware onto it.

To verify that the camera works visit this URL: taunoe.github.io/grove-ai-vision. Use Google Chrome or Edge browser. And click Connect button. Now you should see the camera image.

3.2 Capturing images

Next download the python script capture_images_script.py.

You may need to install some python modules:

pip3 install libusb1
pip3 install pillow
pip3 install opencv-python

Run it capture_images_script.py --interval <time_in_ms>

python3 capture_images_script.py --interval 2000

This firmware takes pictures with a size of 640x480px. This is enough because the models use images of 192x192 px. Although the camera supports 1600x1600 px images. I haven't found firmware or source code for that purpose.

Images will be saved to the folder: save_img/.

The end.