نوع مقاله : علمی-پژوهشی
نویسندگان
دانشکده مهندسی و علوم کامپیوتر، دانشگاه شهید بهشتی، تهران، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Autonomous vehicles have significant potential to improve road safety and traffic capacity by reducing human-related errors while enhancing passenger comfort and welfare. One of the most critical components of these systems is reliable obstacle detection and accurate distance estimation, where any miscalculation may lead to severe human and financial consequences. To improve reliability, autonomous vehicles typically employ heterogeneous sensors such as LiDAR, radar, and cameras. However, adverse environmental conditions can degrade sensing accuracy, and the fusion of heterogeneous sensor data remains a challenging task. In this paper, a novel hybrid framework is proposed to improve the reliability of obstacle detection through health-aware sensor data fusion. The proposed system is evaluated using LiDAR, radar, and camera data for obstacle detection and distance estimation. First, a Long Short-Term Memory (LSTM) network is employed to monitor the health status of each sensor on a frame-by-frame basis. The estimated obstacle distances, together with the detected sensor health conditions, are then fed into an improved adaptive fuzzy fusion module. This health-aware fusion mechanism dynamically adjusts sensor contributions so that the final fused output remains as close as possible to the fault-free ground truth. Experimental results demonstrate that the proposed hybrid framework, which integrates LSTM-based fault diagnosis with adaptive fuzzy fusion, achieves a 22.5% improvement in fusion accuracy, whereas the conventional fuzzy fusion method achieves only 4.1%. This substantial enhancement significantly improves the reliability of obstacle detection systems in autonomous vehicles, enabling safer decision-making and reducing the risk of costly failures.
کلیدواژهها [English]
)