Using specialised cyclist detection software to count cyclists and determine cyclist travel speed from video
Files
(Restricted Access)
Date
2014
Authors
Ponte, G.
Szpak, Z.
Woolley, J.
Searson, D.
Editors
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Conference paper
Citation
Proceedings of the 2014 Australasian Road Safety Research, Policing & Education Conference, 2014, pp.1-13
Statement of Responsibility
Ponte, G., Szpak, Z. L., Woolley, J. E., Searson, D. J.
Conference Name
2014 Australasian Road Safety Research, Policing & Education Conference (ARSRPE 2014) (12 Nov 2014 - 14 Nov 2014 : Melbourne, Vic)
Abstract
The Australian Centre for Visual Technologies (ACVT) developed software for post-processing video footage that is capable of detecting, counting and assessing the level of conspicuity of cyclists. The initial version of the software, on average, correctly identified and tracked 69% of cyclists in footage of busy intersections and roundabouts when first trialled. A number of additional trials were conducted to extend the features of the software. The second trial was undertaken to explore the possibility of automating speed detection. The video detection software correlated well with the true cyclist counts and speeds measured by GPS. The third trial involved recording cyclists travelling over specialised bicycle detection counters and measuring their speeds with a laser gun. This enabled a comparison between the counts provided by the video detection software and the counts provided by the closed induction loop counters as well as a comparison of speeds. The final trial involved four real world sites at which video recordings were taken and analysed by an improved version of the software and compared to human observations. The improved version of the software was able to detect 89 to 98% of cyclists. The results indicated good correlation with human observations and demonstrated the feasibility of using readily obtainable video footage to collect objective bicycle data. This paper briefly summarises the development and improvement of the software, details the methods used to obtain the experimental data, present the results and discusses potential future applications of the software and improvements in detection accuracy.
School/Discipline
Dissertation Note
Provenance
Description
Access Status
Rights
Copyright status unknown