Autonomous Steering Control using AI-Based Driver Drowsiness Detection and Safe-Zone Navigation

Abstract

Driving safety is one of the most elusive issues in the contemporary transportation system across the world, and driver drowsiness and the loss of situational awareness have been cited as the main cause of road accidents. The latest developments in artificial intelligence (AI), computer vision, and sensor fusion have opened opportunities to the further improvement of vehicle autonomy and the minimization of risk events, that are caused by humans. This paper will involve a combined autonomous-steering control structure to relate real-time driver drowsiness to intelligent safe-zone transmission to maintain a balanced operation of the vehicle and safety of the occupants in extreme fatigue-based circumstances. The suggested system utilizes a multimodal framework that uses convolutional neural networks (CNNs), facial-behavioural surveillance, automobile telemetry, and LiDAR-supported environment scanning to identify the signs of driver drowsiness at an early stage. The indicators which include eye closure score (PERCLOS), yawning, deviation in head-pose, and changes in blink rate are key indicators that are analysed using hybrid CNN-LSTM pipeline that is optimized to interpret temporal behaviour. When the state of drowsiness is met, the autonomous intervention module will prompt an AI-based steering controller that will partially take the control of the vehicle. It uses the steering control subsystem based on a Safe-Zone Decision Engine (SZDE) that computes the best approach routes to safety areas that are defined by roadside shoulders, emergency bays, or low deceleration zones. Trajectory tracking is executed by performing real-time dynamic path planning, a combination of search method Astar, smooth Bézier curves and Model Predictive Control (MPC). This system includes a probabilistic risk model with collision probability and environmental uncertainty that quantifies LiDAR, radar, and camera measurements to implement safe control in environments that are partially obstructed or complex. The last important contribution of the work is a synergistic interaction between environmental-aware autonomous navigation and human-state estimation. The system focuses on predictive intervention rather than the reactive correction by detecting the drowsiness onset up to 7-10 seconds before full cognitive degradation takes place. Simulations in CARLA, hardware-in-loop tests as well as real world driving in controlled tracks were also used as experimental evaluations. According to the findings, the autonomous steering system with AI successfully covered the latter by cutting lateral deviation by 62 percent, reducing safe-pull-over time by 34 percent, and keeping the drowsiness classification correct 98.1 percent in a range of illumination conditions. According to the findings, there are high potentials of its use in commercial cars, over the road, long distance freight and consumer level advanced driver-assistance systems (ADAS). Future research incorporates V2X communication to cooperatively negotiate safe zones with adaptive multimodal sensing with thermal imaging and infrared imaging. Comprehensively, the study confirms that combining driver-state observation with autonomous steering control is an encouraging direction of improving road safety and increasing the development of semi-autonomous transportation systems.