sajeevsingh / custom-object-detection-using-open-cv

Train your model for object detection using customized data created by you for that particular object

README FILE

Hey!! It's a super easy method by which you can train your own model for any object you want and use that trained model on live video/downloaded video and image using opencv

STEPS TO FOLLOW

1- Use the 'main.py' file and run in on your pyhton IDE for fetching the images using your web-cam (Note - You can also use any other device to capture the images manually)

2- Run the program two times for obtaining positive and negative images

• Positive - The actual images of the object
• Negative - Images of object similar to the original object or related to it yet false images( Example- For a car model the negative images can be road,traffic signal or anything like that which might come in the detecing videos)

(Note - Always use 2 to 3 times more negative images than the positive images)

The images will be saved and more the number of images from different angles more will be the accuracy

---

3- Now for the implementation part use This website Upload the positive and negative images in the trainer GUI and tell the negative image count and in common set the No. of stages as 17 and in Cascade the Sample width as 38 Then press the START button

4- Download the xml file of the model obtained and then open the 'Using Trained Model.py' and replace the name in the required place and test the modelon any video,image or live video

NOTE- RASBERRY PI CAN ALSO BE INTEGRTED IN THIS

Contact me on - Linekdin