Research Papers

Automated Analysis of Cyclists Safety Analysis and Data Collection using Computer Vision

Version 1
Date added June 18, 2014
Downloaded 0 times/fois
Category 2014 CMRSC XXIV Vancouver
Tags Research and Evaluation, Session 3C
Author/Auteur Mohamed H. Zaki, Tarek Sayed
Stream/Volet Research and Evaluation

For Slideshow Presentation please contact Dr. Tarek Sayed (tsayed@civil.ubc.ca)(no paper submitted)

Abstract

Maintaining a safe cycling environment is vital for a robust transportation system yet it represents several challenges for promoting sustainable travel modes. Non-motorized modes of travel such as cycling may expose the road-user to severe consequences, when involved in traffic collisions). Therefore, a thorough understanding of cyclists’ behavior is of considerable interest.  Emergent sensors technologies in intelligent transportation systems are the vanguard of the means to extend the depth and wealth of information on the road-users and traffic conditions that can be captured. One approach that holds considerable potential for data collection is computer vision analysis.This is a non-intrusive method and can be conducted at a distance. In this paper, we present recent advances in the development of computer vision framework for data collection and safety evaluation of cycling facilities. We tackle the problem of cyclists’ data collection in dense traffic from the perspective of automated counting, density estimation and speed calculation. We also study the cycling behavior among different populations (e.g., gender, fitness level). We present a novel classification procedure that automatically identifies cyclists from other road-users like motorcyclists, vehicles and pedestrians. The classification is based on semi-supervised clustering procedure applied on features extracted from the speed profiles (e.g., cadence frequency) and geometric features (i.e., estimate size) of the road-users. Traffic conflicts based proactive safety evaluation of cycling facilities is also demonstrated as a way to study the interaction between cyclists and other road-users. In this approach, traffic conflicts are automatically detected and their severity ranked using the Time to collision safety indicator. As well, vehicle violations such as failure to respect yield signs are automatically identified. The presented research is applied, evaluated and validated on datasets collected across greater Vancouver, BC, with good results that show the usefulness of the framework. This encourages the applications to wider and large scale data sets. Transportation applications that benefit for this research include simulation models, planning and management of cyclists’ trips and the development of safety and collision prevention programs.

Mohamed H. Zaki, Tarek Sayed