Ling Zhao

Ms. Ling Zhao


Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
12 October 2026 in Smart Cities
Verified
Journal Article
Open access
PDF file
LZLing Zhao
Ling Zhao
Jerry GaoJerry Gao
Jerry Gao
YMYukta Mehta
Yukta Mehta
AAAishwarya Ashok
Aishwarya Ashok
DDDeepthi Desharaju
Deepthi Desharaju
SMSowmya Manchikanti
Sowmya Manchikanti
Rapid urbanization and persistent road-safety challenges have increased the demand for intelligent traffic management systems capable of supporting traffic monitoring, incident analysis, and short-term forecasting. Existing artificial intelligence solutions frequently address these functions through separate, task-specific applications, making it difficult for traffic operators to access heterogeneous analytical services within a common operational environment. This study presents a cloud-based, multi-source traffic intelligence platform that incorporates independent analytical services for CCTV video, UAV video, and IoT sensor data. The proposed platform performs platform-level service integration among the three sources. Its principal analytical component is a multi-model, multi-task UAV video framework consisting of three task-specific pipelines for accident detection and classification, emergency-vehicle detection, and vehicle tracking, counting, and congestion analysis. The semantic outputs of these pipelines are integrated at the decision level to generate annotated videos, traffic statistics, congestion assessments, incident alerts, and consolidated traffic reports. YOLOv8-based models are used for traffic-object and incident detection, Deep SORT is employed for multi-object tracking, and DCRNN is independently applied to PeMS-BAY sensor data for short-term traffic forecasting. A graph-partitioning strategy with overlapping boundary nodes is also evaluated to investigate the trade-off between forecasting accuracy and computational efficiency. Module-level experiments demonstrate effective task-specific performance. The CCTV accident detection model achieves a validation precision of 89.09%, recall of 88.10%, and mAP of 88.09%. The augmented UAV accident-analysis model achieves a validation precision of 82.20% and mAP of 80.20%. For five-minute traffic forecasting, the DCRNN service obtains an MAE of 0.85, MAPE of 1.63%, and RMSE of 1.54 on the PeMS-BAY dataset. These results demonstrate the feasibility of integrating independently processed traffic-analysis services within a common cloud platform while using multi-model UAV analytics to provide consolidated aerial traffic intelligence.

2 June 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
LZLing Zhao
Ling Zhao
Jerry GaoJerry Gao
Jerry Gao
YMYukta Mehta
Yukta Mehta
AAAishwarya Ashok
Aishwarya Ashok
DDDeepthi Desharaju
Deepthi Desharaju
SMSowmya Manchikanti
Sowmya Manchikanti
Rapid urbanization has increased the demand for intelligent traffic management systems capable of real-time monitoring and predictive analysis. Modern smart city applications require the integration of heterogeneous data sources to improve traffic efficiency and safety.However, existing approaches frequently treat vision-based perception and sensor-driven forecasting as distinct processes, limiting their ability to capture dynamic traffic interactions and large-scale spatio-temporal dependencies.To address this limitation, we propose a unified multimodal traffic intelligence framework that integrates YOLOv8-based object detection, DeepSORT-based tracking, and DCRNN-based traffic forecasting, enabling joint perception and prediction within a single system.The framework consists of three tightly coupled modules. First, a vision module performs real-time detection of vehicles and accidents from CCTV and drone data. Second, a tracking module establishes temporal consistency by associating objects across frames. Third, a DCRNN-based forecasting module models traffic flow using IoT sensor data, capturing both spatial and temporal dependencies.In addition, a multi-cluster strategy with overlapping nodes is introduced to enhance scalability while preserving spatial dependencies within large-scale traffic networks. Experimental results demonstrate strong performance across tasks. The detection model achieves up to 93.60% precision and 93.91% mAP, while the integrated accident detection framework maintains robust generalization. The DCRNN model achieves low prediction errors, particularly for short-term forecasting.