Optimized Vehicle Detection Using a Hybrid Deep Learning Model and Edge Intelligence for Embedded Systems

Vehicle detection and image segmentation are essential components of autonomous driving, as they facilitate secure navigation by facilitating precise object recognition and scene comprehension. Although object detection models based on deep learning, such as YOLO and DETR, have substantially improved real-time detection accuracy, there are still obstacles in the areas of detecting small objects, managing occlusions, and optimizing computational performance. This study introduces a hybrid deep learning approach that enhances the robustness and efficacy of detection by incorporating a Detection Transformer (DETR) and YOLOv8 within an Edge Intelligence Framework. The proposed model optimizes real-time performance by utilizing YOLOv8's rapid object localization and DETR's hierarchical attention mechanisms to achieve superior feature extraction. Furthermore, edge intelligence enhances real-time detection capabilities, reduces latency, and distributes processing workloads to optimize computational efficiency. Experimental evaluations on benchmark datasets, including KITTI, COCO, and UAVDT, demonstrate that the hybrid model outperforms standalone object detection models in accuracy, precision, recall, and processing speed. The method significantly improves real-time performance in complex road environments, efficiently manages occlusions, and enhances tiny object detection. Under a variety of environmental conditions, the combination of DETR's attention-based learning and YOLOv8's rapid object localization guarantees accurate detection. Additionally, the model is extremely suitable for embedded autonomous driving systems because the edge intelligence framework reduces computational overhead. The results confirm that the proposed method obtains superior detection accuracy and efficiency, thereby offering a practicable solution for autonomous driving applications.