Workshop on the Comparison of Surrogate Measures of Safety Extracted from Video Data

TRB 93rd Annual Meeting (2014)

Call for Participation

Sponsoring Committee: Safety Data, Analysis and Evaluation (ANB20)

Co-Sponsoring Committee: Artificial Intelligence and Advanced Computing Applications (ABJ70)

Call Description: With the advent of powerful computer vision techniques, video data can be automatically analyzed for an increasing number of transportation applications, including for road safety diagnosis based on surrogate measures of safety that do not require to wait for accidents to happen. While several methods exist for different purposes and settings, few direct comparisons have been made and no guidelines exist to choose and adjust existing methods for a given application. An important reason is the lack of public datasets and of comparison of state of the art methods on tasks relevant for transportation applications (benchmarking). Public datasets and benchmarking are common in several scientific fields, most notably in computer vision (such as the series of IEEE International Workshop on Performance Evaluation of Tracking and Surveillance (PETS) , with 15 workshops from 2000 to 2013) but not in transportation. There is in fact a lack of public data relevant for transportation applications: some general computer vision datasets can be useful, but cover few real life transportation applications.

A group of researchers from the University of Lund, McGill University and Polytechnique Montréal have decided to create such a public video dataset, in particular for road user behaviour analysis and road safety diagnosis. We invite the researchers and practitioners to present the current state of their methods for video analysis, behaviour and safety diagnosis, whether previously published or not. We encourage the workshop participants to test their tools on the videos from the public dataset so that the performance of the different system can be directly compared.

The workshop will take place at the next TRB annual meeting on Sunday afternoon, January 12th 2014 (session 162, 1:30pm- 4:30pm in Marriott, Madison A). Everyone is welcome to attend and participate to the discussions.

Two traffic video datasets are available:

They are available through the Lund University data exchange website, with details about the meta-data and the PDTV software provided to access data and meta-data (PDTV documentation). Software for video analysis and access to the Montréal data is also available in the open source Traffic Intelligence project.

For more information please contact Nicolas Saunier (nicolas.saunier [at] or Aliaksei Laureshyn (aliaksei.laureshyn [at]

Subject: video analysis, surrogate measures of safety, benchmarks, performance metrics

Organizers: Hakan Ardo and Aliaksei Laureshyn (Lund University), Luis Miranda-Moreno (McGill University), Nicolas Saunier (Polytechnique Montreal).