Lane Keeping System using Convolutional Neural Network for Autonomous Car

This study examines supervised learning contributions to self-driving automobile control. The main objective was implementing an end-to-end learning strategy for steering angle prediction from camera data to maintain lane adherence. The study tests the model using a simulated dataset replicating unpredictable vehicle movements and steering issues. The research includes a straight road, a road with four bends, and a U-turn road with four bends and a U-turn. The model's performance is tested in AirSim at 5, 6, and 7 m/s. The initial model successfully prevented lane deviation at 5 m/s on straight highways, bends, and U-turns with 100% success. At 6 m/s, the model achieved 100% lane adherence on straight and bend roads. In U-turn bend scenarios, it had a 98% success rate. At 7 m/s, the model had 100% success in preventing lane deviation on straight and bent roads. In U-turn bend scenarios, the success percentage was 88%. The CNN-based Lane Keeping Assist System (LKAS) model successfully navigated 293 obstacles in a virtual environment simulation at speeds of 5, 6, and 7 m/s, with a success rate of 97.6%. After examination, the LKAS model had 100% success in overcoming 300 barriers at these speeds. This study shows that supervised learning can help self-driving cars maintain lane adherence in various driving conditions. It proves the CNN-based LKAS model's obstacle to negotiating effectiveness in virtual environments.