International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 08 | Aug 2024
p-ISSN: 2395-0072
www.irjet.net
Integrated System for Self-Driving Cars Sam Selvaraj1, Archisha Srivastava1, Rajat Saboo1, Avani Bhuva2 1BTech Computer Science Student, NMIMS MPSTME, Mumbai
2Assistant Professor, Dept. of Computer Engineering, NMIMS MPSTME, Mumbai
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Road safety in India is a pressing concern due to
pose challenges for human drivers and automated vehicles alike. In congested urban settings, the accurate identification of lanes becomes even more critical, as vehicles must navigate through complex traffic scenarios, including intersections, lane merges, and pedestrian crossings.
the prevalence of narrow roads, unclear lane markings, and unrepaired potholes, which frequently lead to accidents and traffic congestion. This issue is further compounded by pedestrians sharing the roads. In response to these challenges, we introduce a Vehicle Lane Detection system that encompasses lane, pothole and pedestrian detection.
The foundation of our system lies in the robust lane detection module, which utilizes Sobel filters and Canny edge detection algorithms to accurately delineate lane boundaries in varying lighting and weather conditions. By extracting relevant features from the road surface, this module serves as the cornerstone for subsequent analyses, enabling precise localization and tracking of the vehicle within its lane. Historically, lane detection predominantly relied on classical computer vision techniques. These methods utilized color and edge-based features to identify lane markings. However, these approaches faced limitations in coping with low-light conditions, inclement weather, and diverse road surfaces.
This paper presents a thorough investigation of the creation and application of a multipurpose vehicle assistance system intended to improve driver experience and road safety. Our system integrates advanced computer vision techniques for lane detection, pedestrian recognition, and pothole detection, culminating in a cohesive solution aimed at real-time analysis of the road environment. The lane detection module utilizes Sobel filters and Canny edge detection, demonstrating robust performance in accurately delineating lane boundaries under varying environmental conditions. Object detection capabilities are facilitated by the MobileNet SSD architecture pre-trained on the COCO dataset, enabling efficient identification of pedestrians and other potential hazards. Pothole detection is addressed through the integration of YOLOv7 pre-trained on the COCO dataset, offering high precision in detecting road surface irregularities. This paper also recommends the development of a live pothole database which can be used by roadway authorities for the efficient management and repair of roads. The proposed system is further validated through integration into consumer-grade hardware, showcasing its computational efficiency and practical feasibility for real-world deployment. Experimental results demonstrate the effectiveness and reliability of the system in enhancing driver awareness and safety on the road.
In addition to lane detection, our system incorporates advanced object detection capabilities facilitated by the MobileNet SSD architecture pre-trained on the COCO dataset. This module enables the real-time identification and tracking of pedestrians, vehicles, and other objects of interest, allowing drivers to anticipate potential collision risks and navigate safely through congested traffic environments. Furthermore, we address the critical issue of road surface maintenance and safety by integrating a pothole detection module based on the YOLOv7 architecture pre-trained on the COCO dataset. This module enables the automated identification and characterization of road surface irregularities, such as potholes and cracks, empowering drivers and road maintenance authorities to take proactive measures to ensure road safety and infrastructure integrity.
Key Words: Lane detection; Pedestrian detection; Pothole detection; Yolo; Mobilenet SSD; Deep learning; Canny edge detection
In essence, this research contributes to the advancement of intelligent transportation systems by offering a comprehensive and robust solution for road environment analysis and hazard mitigation. By combining traditional image processing techniques with deep learning-based object detection models, our system represents a significant step towards safer and more efficient road transportation systems.
1. INTRODUCTION Lane detection plays a pivotal role in ensuring road safety, enhancing driving efficiency, and enabling self-driving technologies. This technology addresses the critical challenge of accurately identifying and tracking lane markings on roadways, providing essential guidance for vehicles to maintain proper lane positioning, make informed decisions, and avoid collisions.
2. LITERATURE SURVEY 2.1 Lane Detection
Owing to the intricate and ever-changing nature of road conditions, lane identification systems must be strong and dependable. Narrow roads, variable lighting conditions, adverse weather, and the presence of multiple lanes often
© 2024, IRJET
|
Impact Factor value: 8.226
Lane detection is a vital component of sophisticated driver assistance systems and driverless vehicles. A literature survey on lane detection techniques reveals a diverse range
|
ISO 9001:2008 Certified Journal
|
Page 363