Mohamed-Rag/Autonomous_Vehicles

Egypt-Aware Autonomous Driving Object Detection built by enhancing BDD100K with real Egyptian road scenes. Features a fine-tuned real-time YOLOv11s model that reliably detects local vehicles (like tuk-tuks), pedestrians, lane violations, and visual noise such as banners optimized for chaotic urban traffic in Egypt. Includes modular scripts for data # Egypt-Aware Object Detection for Autonomous Driving ## 1. Project Overview This project focuses on developing a robust object detection model specifically tailored for the unique and often chaotic road environments of Egypt. By enhancing the BDD100K dataset with local Egyptian road images, we fine-tuned a YOLOv11s model to reliably detect objects, including local anomalies like tuk-tuks and visual noise such as banners, which are critical for autonomous driving and traffic monitoring systems in the region. ## 2. Team Information | Name | | :--- | | Mohamed Ragab Abdelhamid | | Mohsen Ibrahim Hasan | | Fatima Waleed Mostafa | | Ebram Magdy Adolf Ibrahim | | Ramy Mohsen Abdelmoneim | | Ahmed Mohamed Abdelmoneim | ## 3. Problem Definition Most existing autonomous driving datasets fail to capture the **chaotic and unique nature of Egyptian roads**. This includes: * The unexpected presence of **tuk-tuks**. * **Pedestrians** crossing randomly. * Frequent **lane violations**. * **Visual noise** such as banners and non-standard signage. This lack of representation leads to poor performance and reliability of off-the-shelf object detection models when deployed in Egypt. ## 4. Proposed Solution We developed a custom **YOLOv11s** model trained on the BDD100K dataset, which was significantly enhanced with locally-captured Egyptian road images. This enhancement specifically targeted underrepresented classes and scenarios, including tuk-tuks, banners, and realistic lane-violation situations. The resulting model demonstrates strong performance and superior generalization in real Egyptian environments. ## 5. Dataset and Preprocessing * **Base Dataset**: BDD100K. * **Enhancement**: Extended with locally-captured Egyptian images to improve detection of local objects (e.g., tuk-tuks, banners). * **Preprocessing Steps**: * Removal of unlabeled and duplicate images. * Ensuring synchronization between images and labels. * Application of class-specific bou …