Video vehicle detection at signalised junctions: a simulation-based study.
Presented at: TRL Conference MOVA and Adaptive Traffic Signal Control, Birmingham, UK.
Many existing advanced methods of traffic signal control depend on information about approaching traffic provided by inductive loop detectors at particular points in the road. But analysis of images from CCTV cameras can in principle provide more comprehensive information about traffic approaching and passing through junctions, and cameras may be easier to install and maintain than loop detectors, and some systems based on video detection have already been in use for some time. Against this background, computer simulation has been used to explore the potential of existing and immediately foreseeable capability in automatic on-line image analysis to extract information relevant to signal control from images provided by cameras mounted in acceptable positions at signal-controlled junctions. Some consequences of extracting relevant information in different ways were investigated in the context of an existing detailed simulation model of vehicular traffic moving through junctions under traffic-responsive signal control, and the development of one basic and one advanced algorithm for traffic-responsive control. The work was confined as a first step to operation of one very simple signalcontrolled junction. Two techniques for extraction of information from images were modelled - a more ambitious technique based on distinguishing most of the individual vehicles visible to the camera, and a more modest technique requiring only that the presence of vehicles in any part of the image be distinguished from the background scene. In the latter case, statistical modelling was used to estimate the number of vehicles corresponding to any single area of the image that represents vehicles rather than background. At the simple modelled junction, each technique of extraction enabled each of the algorithms for traffic-responsive control of the signals to achieve average delays per vehicle appreciably lower than those given by System D control, and possibly competitive with those that MOVA would give, but comparison with MOVA was beyond the scope of the initial study. These results of simulation indicate that image analysis of CCTV pictures should be able to provide sufficient information in practice for traffic-responsive control that is competitive with existing techniques. Ways in which the work could be taken further were discussed with practitioners, but have not yet been progressed.
|Type:||Conference item (Presentation)|
|Title:||Video vehicle detection at signalised junctions: a simulation-based study|
|Event:||TRL Conference MOVA and Adaptive Traffic Signal Control|
|Open access status:||An open access version is available from UCL Discovery|
|UCL classification:||UCL > School of BEAMS > Faculty of Engineering Science > Civil, Environmental and Geomatic Engineering|
Archive Staff Only