Jul 2026· Indian Journal of Science and Technology· Vol 19, pp. 1955-1963· 0 citations
TL;DR
The proposed work introduces adaptive frame slicing scheme in the input data loader and geometric positional encoding which enables the detection of faraway vehicles with high accuracy in wide area surveillance imagery.
Abstract
Objectives: To develop a reliable multi-vehicle detection-based tracking framework under unregulated traffic scenario using enhanced vision transformer model that can precisely localize and track the vehicles. Method: The proposed framework involves a three-level enhancements, (1) Adaptive frame slicing technique in the input data loader to capture the discriminative features of near and far away vehicles in the video frame (2) Geometric positional encoding to provide clear spatial cues and fusion of features, allowing the model to better separate and detect multi-scale vehicles (3) Dual inferencing of sliced and full frame with DIoU-NMS post-processing technique to remove redundant detections. The proposed framework is evaluated on our custom developed Active Learning based vehicle dataset, “AU-INV-P-PALS”, and two public traffic datasets- “IITM-HeTra and Udacity-Fixed small”. Findings: In comparison with the existing baseline DETR model, the proposed model with SGD optimization showed 9.2% increase in the average precision for IoU = 0.5:0.95 indicating its precise localization ability. The model localizes vehicles that occupy as little as 0.2% of the surveillance video frame area. The observed tracking accuracy on real-time traffic CCTV videos indicates that the model can adapt to unregulated traffic environment and achieved appreciable score of about 85.4%, maintaining the tracking stability. The average F1-score of the enhanced DETR obtained for the test frames indicate the generalization ability of the model across different frames. Novelty: The proposed work introduces adaptive frame slicing scheme in the input data loader and geometric positional encoding which enables the detection of faraway vehicles with high accuracy in wide area surveillance imagery. Further, the enhanced model is trained using active learning based custom developed vehicle dataset that significantly reduces the reliance on massive datasets.
Keywords: Deep Learning, Active Learning, Unregulated Road Traffic Control, Vision Transformer
Low-altitude unmanned aerial vehicle (UAV)-based traffic object detection is challenged by substantial scale variations from aerial perspectives, the extremely small pixel proportions of distant traffic participants, complex road background interference, unstable illumination, and severe occlusion in dense traffic scenes. To address these problems, this paper proposes ACP2-YOLO, an improved YOLO11-based detection framework for low-altitude UAV traffic scenarios, with the goal of enhancing the detection of vehicles, pedestrians, and non-motorized traffic participants. The proposed framework introduces two key improvements. First, a lightweight hybrid ACmix module that integrates convolution and self-attention is embedded into the network, enabling the model to jointly capture local detailed features and global contextual dependencies and thereby strengthen feature representation under complex backgrounds. Second, a P2 small-object detection layer is added to the original three-scale detection structure of YOLO11 to construct a four-scale P2–P5 feature pyramid. By allowing shallow high-resolution features to directly participate in object prediction, this design effectively reduces spatial information loss caused by deep downsampling and improves small-object perception. Experiments on the VisDrone2019 dataset show that the improved model achieves 53.1% Precision, 41.1% Recall, 42.9% mAP@50, and 26.3% mAP@50–95, outperforming the baseline YOLO11 by 4.2, 4.2, 5.0, and 3.6 percentage points, respectively. Comparisons with mainstream YOLO-series detectors further demonstrate its superior overall accuracy, small-object detection capability, and adaptability to complex scenes, indicating its potential for UAV-based traffic monitoring, road safety inspection, and intelligent transportation perception.
Tian Luan, Fan Yang, Huan-Xia Wei et al.· Mathematics· 0 citations
To address the challenges of construction vehicle detection caused by complex illumination and severe occlusion in transportation infrastructure scenarios, as well as the requirements for emergency rescue under sudden hazards, this paper proposes a detection system based on a multi-model decision fusion architecture. The system adopts a “coarse-screening and fine-inspection” strategy, utilizing YOLOv8 for rapid global screening to ensure real-time performance. For regions with low confidence, an improved DPFL-RetinaNet is introduced to perform secondary refined detection. Concurrently, this paper proposes the Decoupled Polynomial Focal Loss based on Taylor series expansion, which enhances the mining capability for weak targets through gradient reshaping. Experimental results demonstrate that the proposed system achieves an AP50 of 87.1% on the TCVD dataset, representing an improvement of nearly 30 percentage points over the standalone YOLOv8. This provides an efficient visual perception solution for construction safety and emergency rescue in complex environments, effectively balancing detection precision and inference speed. The proposed Confidence-Guided Cascade Decision System (CG-CDS) effectively addresses the inherent trade-off between computational efficiency and detection accuracy. By dynamically routing challenging samples to a specialized fine-grained detector, the system achieves robust performance across diverse environmental conditions including low illumination, severe occlusion, and complex background clutter. Extensive ablation studies confirm that each component of the cascaded architecture contributes positively to the overall detection performance, with the proposed Decoupled Polynomial Focal Loss (DPFL) providing the most significant improvement for weak target mining.
Urban traffic areas, such as intersections, significantly escalate the risk of accidents due to the high density and diverse behaviour of road users. Achieving higher automation levels necessitates robust object recognition and tracking capabilities to mitigate accident risk. Modern vehicle architectures utilize multiple sensor systems to ensure robustness and high-quality perception. For camera systems in particular, multi-camera multi-target tracking (MCMTT) with visual re-identification (ReID) is fundamental for understanding object behaviour across different fields of view. This work directly addresses this need by presenting and evaluating a deep learning-based concept for near-field MCMTT using Waymo’s real-world perception dataset [1]. This work enables comprehensive and reliable tracking, focusing on the investigation of concept parameters and highlighting the essential factors for the automated driving task.
Marius Westendorf, Jonas Brinkmann, Marcel Kascha et al.· Automotive and Engine Techno...· 0 citations
Vehicle detection and tracking in unmanned aerial vehicle (UAV) imagery, while critical for intelligent transportation systems, remain challenging due to high omission rates, false alarms, and frequent identity switches among small-sized vehicles. The proposed research establishes an enhanced tracking-by-detection framework integrating an improved YOLOX with DeepSORT to mitigate the aforementioned challenges. Integrating the Convolutional Block Attention Module (CBAM) into the feature fusion stage optimizes feature extraction within complex environments. Furthermore, the detection head is refined by replacing standard Binary Cross-Entropy (BCE) and IoU losses with Varifocal Loss and Enhanced IoU (EIoU) loss, respectively, to mitigate sample imbalance and boost localization accuracy. A sliding-window-based image slicing method is also introduced to enhance detection sensitivity to small-sized spatial features. Experimental results demonstrate that the proposed method significantly reduces detection errors and identity switches while strengthening tracking stability.
Jianping Zeng, Jiang-Hong Zhu· International Conference on...· 0 citations
Accurate vehicle localization from monocular roadside surveillance cameras is an important problem in intelligent transportation systems, traffic monitoring, and traffic conflict analysis. Standard localization approaches typically estimate vehicle position using the center of the detector bounding box, which may lead to large localization errors due to perspective distortion and parallax effects, particularly for elevated roadside cameras and large vehicles.This paper proposes a two-stage geometry-aware localization pipeline that estimates the projection of the vehicle footprint onto the road plane instead of relying directly on detector geometry. In the first stage, vehicles are detected using a YOLO26-based detector. In the second stage, a dedicated ResNet34 regression network predicts four corner points corresponding to the projection of the vehicle base onto the image plane. The final vehicle position is estimated as the geometric center of the predicted quadrilateral.The proposed method was trained using synthetic data generated in the CARLA simulation environment and subsequently fine-tuned on real-world roadside imagery from the DAIR-V2X dataset. Experimental evaluation performed on both synthetic and real-world data demonstrated clear improvements in localization accuracy compared with naive bounding-box-center-based localization. On the DAIR-V2X dataset, the proposed approach reduced the mean image-space localization error from 31.77 px to 15.30 px (51.8% improvement) and the median error to 4.29 px. Median ground-plane localization error for medium-range vehicles decreased from 5.52 m to 0.90 m, while for far-range vehicles it decreased from 8.67 m to 1.84 m.The experiments additionally demonstrated that contextual information surrounding the detector bounding box plays an important role in geometric localization. The largest improvements were observed for distant vehicles and geometrically challenging cases affected by strong perspective distortion and parallax effects.
Jan Gawroński, W. Czajewski· 2026 Progress in Applied Ele...· 0 citations
Rapid global urbanization is increasing the need for accurate, large-scale traffic monitoring to support sustainable transportation and city governance. Satellite video remote sensing offers a unique way to continuously observe urban road networks over large areas. It provides high-resolution spatio-temporal data that is essential for traffic flow analysis, infrastructure assessment, and dynamic urban planning. Moving vehicle detection in satellite video sequences is a basic task that turns raw imagery into useful traffic-state information, supporting these applications. Despite the advantages of satellite video data, detecting moving vehicles in practice remains a tough problem. Objects are extremely small and lack clear appearance details, while low local contrast makes them hard to separate from complex backgrounds. Satellite platform motion also introduces background misalignment and intensity fluctuations, resulting in missed detections and false alarms that hurt monitoring reliability. Furthermore, current methods do not fully exploit temporal motion cues or transform-domain priors, creating a performance bottleneck that restricts their practical use. To solve these problems, this paper proposes a Channel-Attentive Spatio-Temporal-Frequency Alignment (CASTFA) framework to effectively use and combine multi-dimensional features for moving vehicle detection in satellite videos, with the goal of providing high-quality traffic monitoring data to help smart city planning. Specifically, a State Space-Guided Temporal Compression (SSGTC) module first collects information along the time dimension with linear computational complexity, greatly reducing overhead while keeping motion cues that are critical for traffic-state estimation. The compressed temporal features are then processed with a multi-scale Haar wavelet transform to get hierarchical time-frequency representations that capture subtle motion dynamics across different frequency bands. At the same time, a pre-trained backbone network extracts multi-scale spatial features. To allow these different domains to work together, a Cross-Domain Feature Alignment (CDFA) mechanism aligns and combines spatial and time-frequency features through channel-attentive operations. Experimental results on the publicly available satellite video moving vehicle detection dataset show that the proposed CASTFA method consistently outperforms existing approaches, with better precision, recall, and F1-scores across diverse urban scenarios. These results show that CASTFA can provide reliable moving vehicle detection performance under difficult real-world conditions, supporting accurate traffic-flow monitoring and providing valuable geospatial intelligence for smart urban planning, transportation management, and sustainable city development.
Ning Zhao, Xiao Wang, Xiaopeng Zhang et al.· ISPRS International Journal...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
