A Vehicle Detector Based on Feature Focused Diffusion

One-Stage Object Detectors in Autonomous Driving

Autonomous vehicles depend on fast and reliable perception systems to detect surrounding vehicles, pedestrians, cyclists, traffic signs, and other road objects in real time. This paper presents a comprehensive survey and analysis of one-stage object detectors for autonomous driving rather than an implementation of a new detection system. The survey reviews the evolution of major one-stage detectors, including YOLOv1, SSD, RetinaNet, EfficientDet, anchor-free detectors such as FCOS and CenterNet, and recent real-time models such as YOLOv10. It compares these architectures through their design choices, feature-fusion strategies, loss functions, deployment trade-offs, and reported benchmark performance. The paper also summarizes commonly used autonomous-driving datasets, evaluation metrics, open challenges, and future research directions. Overall, this survey highlights how one-stage detectors balance speed, accuracy, efficiency, and robustness, while also emphasizing the remaining gap between benchmark results and dependable real-world autonomous-driving performance.

Improved detection accuracy and efficiency for small-object detection in drone imagery with lightweight LightS-DETR

Knowledge-Driven Multi-Scale Vehicle Detection Framework for Intelligent Transportation Systems

In complex traffic environments, conventional vehicle detection methods often show limited precision and robustness when facing distant, small-scale, and occluded vehicles. To address these issues, this research proposes a Multiscale, Multi knowledge, Attention enhanced You Only Look Once (YOLO) model. It is a knowledge-driven multiscale vehicle detection framework for intelligent transportation systems, built on the YOLO version 8 nano model. The framework introduces three modules. The first is a cross-stage partial mixed aggregation network module. This module enhances backbone representation through dynamic multiscale aggregation. The second is a multiscale downsampling module that combines dilated convolution and parallel pooling to preserve vehicle cues during downsampling. The third module is an aspect-ratio perception with cross-attention module that injects aspect-ratio-aware knowledge with cross-attention to adapt to vehicle shape variation and suppress background interference. Experiments on the Vehicle dataset on Kaggle.com and the Berkeley DeepDrive 100K dataset showed that the Multiscale, Multi knowledge, Attention enhanced-YOLO model outperformed the lightweight YOLO baseline models. On the Berkeley DeepDrive 100,000 dataset, it achieved a 48.17% mean average precision @0.5 and 26.70% mean average precision @0.5:0.95, improving YOLO version 8 nano by 3.34% and 2.19%, respectively. Visualization results further confirmed its effectiveness under occlusion and adverse-weather conditions while maintaining real-time efficiency.

Triple-attention enhanced and RepViT-driven LiDAR 3D object detection for complex traffic scenarios

Improved YOLO11n for traffic light small object detection

To address the challenges of high detection difficulty caused by small target size and complex background environments of traffic lights in urban road scenarios, this paper proposes a YOLO11n-based traffic light detection algorithm, named YOLO11n-PRE. Based on the baseline YOLO11n, the proposed algorithm first replaces the original C3k2 module in the backbone network with the C3k2-RCB module, which enhances deep feature extraction capability while maintaining lightweight via efficient residual connection and feature recalibration mechanism. Secondly, a P2 small object detection layer is introduced to fuse shallow high-resolution feature information, expanding the original 3-scale detection architecture to 4-scale detection, which significantly strengthens the network’s perception and positioning ability for micro-scale traffic lights. Finally, an ECA efficient channel attention module is embedded into the PANet feature fusion path, which enhances key feature response and suppresses redundant information via adaptive weight calibration in the channel dimension. Experimental results show that compared with the baseline YOLOv11n, YOLO11n-PRE achieves a 1.9 percentage point improvement in mAP@0.5, with precision and recall increased by 1.8 and 2.3 percentage points respectively, and the parameter count is only 2.8M, which fully meets the requirements of real-time and efficient detection.

A YOLOv8-Based Real-Time Road Congestion Decision-Making Approach Fused with Channel–Spatial Attention and Dynamic Weighted Loss

Conventional object detection models suffer from significant performance degradation in dense urban traffic scenarios. To address these critical limitations and enable accurate real-time road congestion decision making, this study proposes an optimized YOLOv8-based detection paradigm that decouples multi-scale feature enhancement from dynamic focused bounding box regression. Specifically, a multi-scale feature enhancement (MFE) module is designed to extract high-resolution shallow features directly from the P2 layer of the YOLOv8 backbone. Then, a convolutional block attention module (CBAM) is embedded into the feature fusion neck to adaptively filter complex urban background noise and recalibrate channel–spatial feature responses for vehicle target saliency. Furthermore, the standard CIoU loss is replaced with the Wise-IoU (WIoU) dynamic focusing loss function, which suppresses gradient interference from low-quality, occluded samples and stabilizes bounding box regression for dense vehicle targets. The high-precision vehicle detection outputs are fed into a quantitative congestion index (CI) model, which fuses vehicle density and average speed to realize real-time congestion-level classification. Extensive experiments on the public UAVDT benchmark dataset demonstrate that the proposed model achieves an mAP@0.5 of 83.1% (3.8 percentage points higher than the YOLOv8 baseline), an mAP_S (small target) of 23.2% (a 4.3 percentage point improvement), and a real-time congestion decision accuracy of 83.8%. Ablation studies verify the independent and synergistic effectiveness of the MFE, CBAM, and WIoU modules, with the MFE module making the greatest contribution to small-target detection performance (+1.7% mAP@0.5). The proposed model maintains a real-time inference speed of 86 FPS (frames per second) on an NVIDIA RTX 3090 GPU, far exceeding the 30 FPS threshold for real-time traffic monitoring.