Video Streaming Analytics for Traffic Monitoring Systems

Amin, Muhammad Arslan; Kashif, Muhammad; Umer, Muhammad; Kamran, Muhammad; Kanwal, Ayesha; Azeem, Muhammad · 2018 · Crossref

DOI: 10.14569/ijacsa.2018.091192

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This review paper addresses the inefficiencies of traditional traffic monitoring systems, which rely on manual observation and are costly, time-consuming, and prone to human error. The authors argue that the exponential growth of video data from surveillance cameras necessitates automated, scalable solutions. The study focuses on video streaming analytics for traffic monitoring, specifically examining how Hadoop-based frameworks can efficiently process large-scale video streams to detect objects, classify vehicles, and monitor traffic conditions such as congestion and accidents. The paper reviews several technical approaches for implementing these systems. It first evaluates object detection algorithms, noting that while template matching and background subtraction have limitations regarding new objects or computational cost, cascade classifiers offer a more affordable and effective solution. The core of the review analyzes three Hadoop-based frameworks. The first, proposed by Abdullah et al., utilizes a Hadoop-based GPU cluster to analyze recorded video streams stored in the cloud, enabling faster processing than CPU-based methods. The second framework, by Natarajan et al., employs Hadoop MapReduce for live video analysis. This system splits raw video streams into 64 MB blocks distributed across HDFS nodes, using a mapping phase to index frames and a reducing phase to aggregate data on vehicle speed, accidents, and congestion into Hive tables for further analysis and alerting. The third approach addresses compatibility issues, as Hadoop lacks native video interfaces and is Java-based, while many analytics tools are C/C++. Zhao et al.’s Hadoop Video Processing Interface (HVPI) bridges this gap by converting video inputs into key-value pairs for MapReduce processing. Finally, the paper discusses Chen et al.’s integration of deep learning, specifically using YOLOv2 and Convolutional Neural Networks (CNNs) within the HVPI and MapReduce framework for real-time vehicle detection and license plate recognition. The findings indicate that Hadoop-based architectures significantly enhance the scalability and speed of traffic video analytics. The MapReduce framework effectively handles live data by parallelizing frame processing, allowing for real-time detection of traffic anomalies and automated routing suggestions. The integration of HVPI resolves compatibility barriers, allowing existing C/C++ applications to run on Hadoop. Furthermore, the combination of deep learning algorithms like YOLOv2 with MapReduce provides high accuracy and reduced processing time for vehicle detection and recognition tasks. The significance of this work lies in demonstrating that cloud-based, distributed computing frameworks can replace manual traffic monitoring with automated, accurate, and cost-effective systems. By leveraging Hadoop, GPU clusters, and deep learning, these systems can handle large-scale video data to improve public safety, manage traffic congestion, and provide real-time alerts for accidents. The review highlights the evolution from simple object detection to complex, real-time deep learning applications, establishing a foundation for future intelligent traffic management systems.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• distraction detection algorithms