Advanced Radar Signal Processing Using Deep Learning for Real-Time Object Detection and Tracking in Autonomous Vehicles

Real-Time Vehicle Detection and Location Tracking for Military Vehicles Using YOLOv8 and Multi-Object Tracking

Intelligent transportation systems, traffic surveillance and smart city monitoring require accurate vehicle detection and tracking. Traditional methods of monitoring are usually based on either manual monitoring or GPS-based monitoring which may have issues with signal dependency and lack of scalability. This paper suggests a deep learning-based car surveillance system, which combines the YOLOv8 object detector and a multi-object tracking system to perform automated automobile detection and tracking. A prepared set of custom traffic image data (about 1,700 images) was divided into a training (80%) and validation (20%) sample and trained on the YOLOv8-Nano model to detect vehicles. Images had been resized to 640 × 640 resolution during training with 50 epochs of a batch size of 16 with transfer learning on pretrained weights. The trained detector had the accuracy of 0.81 with the recall of 0.74 and the mean Average Precision (mAP-0.5) of 0.79 on vehicle detection assignments. The detecting unit was further combined with tracking structure to retain vehicle identities between sequential frames to provide the capability of regular tracking of multi-object in traffic environment. The test outcomes show that the suggested system can work at about 40 frames per seconds (FPS) with the evaluation dataset and still retain a good tracking precision of about 0.76. Bounding boxes, tracking IDs and performance graphs are some of the visualization results, which support the efficiency of the methodology. The given framework can also be used to track the location of military vehicles in surveillance domains during the situations when the convoy movements or tactical vehicle location can be monitored automatically in GPS-denied or irregular conditions.

An Infrared and Visible Image Fusion-Based Method for Air-to-Ground Target Tracking

To address the challenges of low illumination, occlusion, complex backgrounds, platform motion, and dense distributions of small-scale targets in UAV-based air-to-ground missions, this paper investigates a target tracking method based on infrared and visible image fusion. Based on the complementarity of dual-modal imaging, MoME-Track is constructed for continuous single-target locking, while DCAF-Net/MAMC-Track is developed for multi-target detection and tracking. For single-target tracking, dynamic collaboration between appearance information and motion priors is achieved through cross-modal appearance experts, an extended Kalman filtering-based motion expert, and a mixture-of-experts decision mechanism. For multi-target tracking, a dual-branch cross-domain fusion detection network is designed to extract infrared and visible features. Frequency-spatial collaborative enhancement and multi-head cross-attention are introduced to improve small target detection capability. In the tracking stage, multimodal appearance measurement, depth-adaptive Kalman filtering, and low-confidence detection reuse are incorporated to enhance trajectory continuity and identity consistency. Experiments conducted on the self-constructed MSOT-UAV and MMOT-UAV datasets, as well as public datasets, demonstrate that the proposed method achieves a favorable balance among accuracy, robustness, and real-time performance.

Autonomous Vehicle Navigation Using Sensor Fusion Algorithms

Autonomous vehicle navigation is a key component of modern intelligent transportation systems, relying on the integration of multiple sensors such as LiDAR, radar, cameras, GPS, and IMUs. Sensor fusion techniques combine data from these sources to improve perception, localization, and reliability. This paper reviews classical pre-2018 sensor fusion methods, including Kalman Filters, Extended Kalman Filters (EKF), Unscented Kalman Filters (UKF), and particle filters. Different sensors have individual limitations—cameras are affected by lighting, LiDAR is costly, and radar has lower resolution—but fusion enhances overall system performance by leveraging their complementary strengths. The study examines low-, mid-, and high-level fusion approaches and proposes a hybrid framework using GPS/IMU for localization and LiDAR-camera fusion for obstacle detection. The system is based on probabilistic and Bayesian models, designed for real-time performance and robustness against noise and sensor failures. Key challenges such as synchronization, calibration, and computational complexity are discussed. Results show that sensor fusion significantly improves navigation accuracy, highlighting the importance of selecting appropriate algorithms based on application needs.Overall, the paper emphasizes that multi-sensor fusion is essential for safe and reliable autonomous driving and provides a foundation for future advancements in the field.

Enhancing Radar PPI-Based Maritime Object Detection Using Hybrid Preprocessing and Deep Learning Approach

In this work, we present a hybrid framework for object detection in maritime radar PPI frames. The method combines domain-specific preprocessing with deep learning models, focusing on YOLOv8s and YOLOv12s. Challenges such as low signal-to-noise ratios, sea clutter, and small target visibility are addressed using adaptive binarization, morphological operations, and noise suppression. A high-resolution P2 feature layer and an area-aware loss function improve small object detection, especially for low-contrast targets such as buoys. The proposed approach is evaluated on 350 simulated radar frames containing annotated fixed objects, moving targets, and landmasses, using five YOLO models (YOLOv5s, YOLOv8s, YOLOv10s, YOLOv11s, YOLOv12s). YOLOv12s achieved the best performance for fixed-object detection on the preprocessed dataset, achieving an mAP50 of 0.772 and an mAP<inline-formula> <tex-math notation="LaTeX">${}_{50-95}$ </tex-math></inline-formula> of 0.503. For the overall evaluation, YOLOv12s obtained the highest mAP50 score of 0.917, while YOLOv8s achieved the best mAP<inline-formula> <tex-math notation="LaTeX">${}_{50-95}$ </tex-math></inline-formula> score of 0.762. After integrating the P2 layer and area-aware loss function, both models produced very similar results. YOLOv8s achieved an mAP50 of 0.957 and mAP<inline-formula> <tex-math notation="LaTeX">${}_{50-95}$ </tex-math></inline-formula> of 0.813, while YOLOv12s achieved an mAP50 of 0.958 and an mAP<inline-formula> <tex-math notation="LaTeX">${}_{50-95}$ </tex-math></inline-formula> of 0.812. In addition, external validation on the real-world DAAN marine radar dataset demonstrated that the gains achieved on simulated data successfully transferred to real radar imagery. The full configuration (preprocessing + P2 layer + area-aware loss) achieved the best performance with mAP50 of 0.922 and mAP<inline-formula> <tex-math notation="LaTeX">${}_{50-95}$ </tex-math></inline-formula> of 0.745, corresponding to substantial improvements over the raw baseline with mAP50 of 0.698, and mAP<inline-formula> <tex-math notation="LaTeX">${}_{50-95}$ </tex-math></inline-formula> of 0.675. These findings indicate promising improvements in the detection of small and low-contrast objects, contributing to reliable real-time situational awareness for autonomous maritime navigation.

UAV-Based Target Detection, Tracking, and Prediction: A Survey From Classical Models to AI-RAN Intelligence

Recent advances in unmanned aerial vehicle (UAV)-based target detection and tracking increasingly rely on model-based techniques such as Kalman filtering (KF) and particle filtering (PF), as well as learning-driven approaches including deep learning and reinforcement learning. Despite these developments, existing UAV systems continue to face major challenges arising from increasing target densities, complex terrain, dynamic wireless conditions, communication limitations, and restricted onboard computational and energy resources. These constraints significantly affect tracking accuracy, real-time responsiveness, and service reliability, particularly in resource-constrained and rapidly changing environments. To address these challenges, this survey presents a systematic and comprehensive review of UAV-based target detection, tracking, and prediction methods, spanning classical estimation models, deep learning frameworks, reinforcement learning strategies, and cooperative multi-UAV intelligence. The survey further emphasizes the integration of UAV-assisted edge computing with the Open Radio Access Network (O-RAN) framework, where the RAN Intelligent Controller (RIC), together with xApps and rApps, enables scalable, low-latency, and adaptive communication optimization between aerial and terrestrial nodes. Building on this foundation, the survey introduces an Artificial Intelligence-Driven Radio Access Network (AI-RAN)-enhanced conceptual framework that combines particle filtering, Q-Learning (QL) control, and AI-driven RAN optimization to enable joint communication, computation, and control. The proposed architectural perspective demonstrates how multi-modal sensor fusion and distributed edge intelligence can jointly improve tracking robustness, responsiveness, and energy efficiency. Finally, the survey highlights open challenges and future research directions toward fully autonomous, scalable, and network-aware UAV tracking systems for emerging 6G and edge-AI environments.

Signal Processing, Optimization, and Multi-Modal Fusion for Next-Generation Automotive FMCW Radar Systems