Computer Vision Recaptcha Solver
Trained and deployed an object detection model to solve google recaptchas and obtained a 52% accuracy improvement versus the random selection baseline.
Overview
This application automates the interaction with a reCAPTCHA challenge found on a webpage, and is specifically designed to handle Google’s reCAPTCHA system. It utilizes the Selenium WebDriver for browser automation, PIL for image manipulation, and a YOLOv5 deep learning model that is retrained and tuned on an extensive hand curated dataset of recaptcha images for the purpose of object detection within the reCAPTCHA images.
Technologies Used
Computer Vision
AWS
Python
Yolo
Selenium
Feature Engineering
Data Collection
Key Features
- Custom dataset curation
- Object detection with Yolov5
- Chrome extention allowing webpage automatic solving
- Performance metrics to evaluate performance and challenges in the model