Project Title: Pakistani Night City Driving Dataset (PNDD)

Description: The Pakistani Night City Driving Dataset (PNDD) is an innovative high-definition dataset meticulously curated to address the challenges of autonomous driving in nighttime urban environments, particularly in regions like Pakistan. This dataset serves as a crucial resource for researchers and developers working on perception systems for autonomous vehicles, enabling them to train and test algorithms specifically tailored to navigate the complexities of nighttime driving conditions.

email: muhammadusamabhatti009@gmail.com For Dataset. Key Features:

Comprehensive Coverage: PNDD comprises over 10,000 high-resolution images captured from various urban environments across multiple cities in Pakistan. These images provide a comprehensive representation of the challenges faced by autonomous vehicles in nighttime driving scenarios. Rich Annotations: The dataset includes detailed annotations for a wide range of objects essential for safe navigation, such as cars, pedestrians, and traffic signs. These annotations facilitate the development and evaluation of perception algorithms for object detection and recognition. Specialized Object Detection Models: The project utilizes state-of-the-art deep learning architectures, with a focus on YOLOv5, optimized for object detection in low-light conditions. These models are fine-tuned using the PNDD to achieve accurate and reliable performance in nighttime urban environments. Visualization Tools: The project includes tools for visualizing dataset characteristics, such as bounding box density heatmaps and instance distribution charts. These visual aids provide valuable insights into the distribution and frequency of different object classes within the dataset.

Methodology:

Data Acquisition: High-definition video data was captured using GoPro Hero 8 cameras mounted on vehicles across various urban and inter-city locations in Pakistan, ensuring a diverse range of driving conditions.

Dataset Preparation: The recorded video data underwent thorough preprocessing, including frame selection, dynamic range adjustment, and noise reduction techniques, to improve visibility and clarity, especially in low-light conditions.

Annotation Process: Each video frame was meticulously annotated using the V7 Image Labeling Tool to define precise boundaries for different objects, ensuring a comprehensive and accurate dataset for training perception algorithms. Object Detection and Recognition: Deep learning models, particularly YOLOv5, were optimized for the PNDD by fine-tuning model parameters to enhance object detection performance in nighttime urban environments.

Usage: Researchers and developers can utilize the PNDD for training and evaluating perception algorithms for autonomous driving systems, particularly those designed to operate in nighttime urban environments. The dataset and accompanying tools can be incorporated into machine learning pipelines for object detection and recognition tasks, facilitating the development of safer and more efficient autonomous vehicles.

Contributing: Contributions to the project, including enhancements to dataset annotations, optimizations of object detection models, and improvements to visualization tools, are welcome through pull requests.