Real-Time Vehicle and Lane Detection using Modified OverFeat CNN: A Comprehensive Study on Robustness and Performance in Autonomous Driving

Abstract

This examination researches the use of profound learning methods, explicitly utilizing Convolutional Brain Organizations (CNNs), for ongoing recognition of vehicles and path limits in roadway driving situations. The study investigates the performance of a modified Over Feat CNN architecture by making use of a comprehensive dataset that includes annotated frames captured by a variety of sensors, including cameras, LIDAR, radar, and GPS. The framework shows heartiness in identifying vehicles and anticipating path shapes in 3D while accomplishing functional rates of north of 10 Hz on different GPU setups. Vehicle bounding box predictions with high accuracy, resistance to occlusions, and efficient lane boundary identification are key findings. Quiet, the exploration underlines the likely materialness of this framework in the space of independent driving, introducing a promising road for future improvements in this field.