richardmiruka / aronayub-custom-data-licence-plate-detection-using-yolo-v8

Train Custom Model using YOLOV8

Custom data sets Licence-Plate-Detection-using-YOLO-V8 + Raspberry Pi

• The main steps involved:

• <> Data logistics
• <> Training the data
• <> Optimizing the model
• <> Integrating with RPI

Data Sources:

• There are two datasets that you can use:

1. 390 images
2. 1001 images

• Open Notebook file and run the commands sequentially.
• Connect to T4 GPU under resources on the right side.

• Under this section: edit the API Key fetched from Roboflow.

!pip install roboflow

from roboflow import Roboflow
rf = Roboflow(api_key="apikeyxxx")
project = rf.workspace("mochoye").project("license-plate-detector-ogxxg")
version = project.version(1)
dataset = version.download("yolov8")

• You can test your model perfomance after training:

 !python /content/Licence-Plate-Detection-and-Recognition-using-YOLO-V8-EasyOCR/ultralytics/yolo/v8/detect/predict.py model='/content/Licence-Plate-Detection-and-Recognition-using-YOLO-V8-EasyOCR/best.pt' source ='directory'

• the source can be:

• Live camera

source = 0

• For images

source = 'directory.png'

• Use whatever images you have to test the model

• video

source = 'directory'

• After training is done, follow the directory to see the test results.

Speed: 0.2ms pre-process, 2.8ms inference, 0.0ms loss, 3.6ms post-process per image
Saving runs/detect/train/predictions.json...
Results saved to runs/detect/train

Integration to RPI

• <> Now that we have a quantized and trained tested model, let's test it in Raspberry pi.

Navigate to the Repository on using YoloV8 with Raspberry Pi

• see how to work with Ultralytics on Raspberry Pi.

• Test the model best.pt

• This can be done for live camera, image and also video.

Challenge for you

• Integrate OCR to get detected data.

• Control GPIO pins after the data is detected.

• Further studies.

Ultralytics Official Link