Comprehensive Review of YOLOv9 and DeepSORT for Vehicle Detection, Multi-Object Tracking, and Counting in Intelligent Transportation Systems
Mr. Saurabh Vyas
CSE (Cyber Security)
G.H. Raisoni College of Engineering and Management Nagpur, Maharashtra, India saurabh.vyas@ghrietn.edu.in
Soham Shrivas
CSE (Cyber Security)
G.H. Raisoni College of Engineering and Management
Nagpur, Maharashtra, India soham.shrivas.cyb@ghrietn.raisoni.net
Rudra Dhoble
CSE (Cyber Security)
G.H. Raisoni College of Engineering and Management
Nagpur, Maharashtra, India rudra.dhoble.cyb@ghrietn.raisoni.net
Tushar Sahu
CSE (Cyber Security)
G.H. Raisoni College of Engineering and Management
Nagpur, Maharashtra, India tushar.sahu.cyb@ghrietn.raisoni.net
Tanuj Panchariya
CSE (Cyber Security)
G.H. Raisoni College of Engineering and Management
Nagpur, Maharashtra, India tanuj.panchariya.cyb@ghrietn.raisoni.net
Saran Surya
CSE (Cyber Security)
G.H. Raisoni College of Engineering and Management
Nagpur, Maharashtra, India saran.surya.cyb@ghrietn.raisoni.net
Amey Bhandarkar
CSE (Cyber Security)
G.H. Raisoni College of Engineering and Management
Nagpur, Maharashtra, India amey.bhandarkar.cyb@ghrietn.raisoni.net
Abstract—Rapid urbanization and motorization have intensi- fied traffic congestion, safety risks, and environmental burdens in cities, making continuous and reliable traffic monitoring a central requirement of modern intelligent transportation systems. Conventional manual analysis of roadside surveillance video is labor-intensive, subjective, and unable to scale with the massive volumes of data generated by ubiquitous CCTV deployments. This review examines the integration of YOLOv9, a recent one- stage object detector, with the DeepSORT multi-object tracker for vehicle detection, tracking, and counting in traffic surveil- lance applications. YOLOv9 introduces Programmable Gradient Information and the Generalized Efficient Layer Aggregation Network, which improve gradient flow, feature reuse, and param- eter efficiency compared to previous YOLO versions. DeepSORT extends classical SORT by combining Kalman filtering and Hungarian matching with deep appearance embeddings to main- tain object identities under partial occlusions and dense traffic conditions. The review discusses the evolution from traditional image processing to deep learning-based detectors, real-time object detection fundamentals, and the progression across YOLO variants. It then analyzes the YOLOv9-DeepSORT pipeline including detection, temporal association, and counting logic, together with metrics such as precision, recall, mAP, MOTA, and FPS. Finally, the paper covers real-world ITS applications, evaluates advantages and limitations, and outlines future research directions.
Index Terms—YOLOv9, DeepSORT, vehicle detection, multi- object tracking, intelligent transportation systems, computer vision, deep learning
