Automated Video Analysis - Analyzing Large Quantities Of Transportation Research Data

NHTSA · 2015 · ROSA P / United States. Federal Highway Administration

archive: archived pipeline: cataloged verified

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

Summary

This paper outlines a research initiative funded by the Federal Highway Administration (FHWA) under its Exploratory Advanced Research (EAR) Program, conducted by the National Robotics Engineering Center at Carnegie Mellon University (CMU). The project addresses the critical challenge of analyzing the massive dataset generated by the Strategic Highway Research Program 2 (SHRP 2) Naturalistic Driving Study (NDS). The NDS collected approximately 2 petabytes of data, including 1.2 million hours of video, from nearly 3,000 volunteers over two years. The primary motivation was the logistical impossibility of manual analysis; extracting features from this volume of video would require approximately 600 technicians working for a full year. Consequently, the research aimed to develop automated tools to reduce the time and cost of data extraction, thereby expanding access to these rich datasets for highway safety researchers. To overcome this data bottleneck, the CMU team implemented a machine-learning-based approach designed to automate feature extraction from complex video data. The methodology utilized algorithms trained on large datasets to identify important features and classify targets with high accuracy. A key innovation involved exploiting contextual cues to interpret ambiguous video data. Rather than relying on traditional graphical models, which can yield inaccurate predictions, the researchers developed a sequence of context-dependent predictions. For example, the system uses inference procedures such as identifying wheels near the corners of a car to classify objects. This approach reduces the need for expert guidance and manages computational effort by leveraging the diversity of the dataset to enhance system robustness. The technical implementation addressed significant data and computational challenges inherent in petabyte-scale datasets. The team developed powerful labeling tools to facilitate fast human analysis and employed anomaly detection to prioritize data access, focusing labeler efforts on significant events. Semi-supervised learning techniques were used to build predictive models while minimizing the amount of labeled data required. To handle the intensive computational load, the researchers explored two strategies: parallelization, which distributes work across multiple network cores, and "anytime predictors," a technique that provides a sequence of results that improve as more analysis time is allocated. The primary outcome of the project is a software framework that supports the efficient application of machine-learning-based feature extraction for large sets of transportation video data. This framework includes specific target detectors capable of identifying both moving targets, such as cars and pedestrians, and static targets, such as traffic signs. The significance of this work lies in its ability to transform the accessibility of the SHRP 2 NDS data. By automating the extraction of details previously difficult to investigate, the project expands the pool of researchers able to utilize these datasets. Furthermore, it advances industry understanding of effective data-processing approaches that can be widely applied to future large-volume transportation data, supporting next-generation transportation solutions.

Key finding

The project successfully developed a machine-learning-based software framework that automates the extraction of features from large-scale transportation video datasets, significantly reducing the manual effort required for analysis.

Methodology

modeling

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. Discovered via bulk_ingest_rosap on 2026-05-23 (7 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

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

• naturalistic crash near crash

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: dataset resource, tool software