rbobjecttracking 2.0.0

A python library for single-object tracking.

rbobjecttracking is based on an algorithm I invented called RBOT. it stands for ROI Based Object Tracking.

To use rbobjecttracking, first create your custom dataset using rbot.dataCollector.

from rbot.dataCollector import DataCollector # Import the DataCollector class

data_collector = DataCollector() # Create the DataCollector object
data_collector.collectData() # This will open a GUI window using customtkinter and opencv.

After filling in the text fields and collecting the image samples, train a small model using rbot.trainer.

from rbot.trainer import Trainer # Import the Trainer class

trainer = Trainer(dataset_path="<dataset name that you created>", batch_size=64, img_size=(128, 128), 
                  lr=<your chosen learning rate>, model_path="<path you want to save your model>") # Create the Trainer object
                  
trainer.train(epochs=<number of epochs you want to train for>) # This will train and save the model to your specified model path

After training, you can then start tracking the object with rbot.rbot.

from rbot.rbot import RBOT # Import the RBOT class
import cv2 # Import opencv for accessing the webcam

rbot = RBOT(hsvValues="blue", image_size=(256, 256), minimum_confidence=0.9) # Create the RBOT object
cap = cv2.VideoCapture(0) # Access the webcam

while True:
    ret, frame = cap.read() # Get a frame from the webcam

    processed_frame = rbot.track_object(frame=frame, color=(0, 255, 0), width=2) # Process the frame

    cv2.imshow("processed frame", processed_frame) # Display the proccesed frame
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

You can even use your GPU for training and inference, if needed. You can still get around 20 FPS on a CPU.

rbot = RBOT(hsvValues="blue", image_size=(256, 256), use_cuda=True) Enable cuda while creating the RBOT object

for training (recommended, but you can still train on CPU)

trainer = Trainer(dataset_path="<dataset name that you created>", batch_size=64, img_size=(128, 128), 
                  lr=<your chosen learning rate>, model_path="<path you want to save your model>", use_cuda=True) # Enable CUDA while Creating the Trainer object