Video Data. Video data were used in this study to validate the IID algorithm. The STAR Lab has access to all the 400+ WSDOT traffic surveillance video cameras deployed along the freeway corridors. These cameras cover I-5, I-405, I-90, SR 167, SR 520, and others. To capture ground-truth travel delay data, two cameras are needed, one at each end of the link being monitored. To facilitate a comparison with loop data-based IID estimates, these cameras’ fields of view are best centered at the loop stations bounding the freeway link. However, very few freeway links have both surveillance video cameras and inductance loops located at approximately the same places. The STAR Lab’s fiber connection can support two live video streams simultaneously, which satisfied the video data collection need for this project. Two video streams, one at each end of a freeway link, were recorded simultaneously to collect vehicle arrival and departure data for the link. Because the research team could not predict when an incident would occur, the video streams were recorded for long periods to ensure the capture of traffic volumes under incident impacts.
Video Data. Video data is essential in understanding or interpreting the data recorded by the sen- sors and in understanding driver behaviour in general. Significant developments con- tinue with regard to cameras, frame rate, resolution, zoom level and storage capacity. Video data analysis will almost certainly be necessary for an ND study that aims to evaluate distraction or inattention in drivers and to study driver behaviour in general. Studying visual behaviour in particular is currently complex and expensive as it re- quires high sensitivity eye-tracking equipment and/or machine-vision techniques. An al- ternative could be the use of event-triggers to highlight areas of interest in the data, such as high longitudinal ‘g’, lateral accelerations or yaw. This can be complemented by manual video analysis to determine the exact nature of the distraction or inattention at specific points of interest. This technique relies on good quality video data in terms of, but not exclusive to, the number of camera and their angles, video resolution and frame rate. For a more complete understanding of both vehicle and environment factors it is rec- ommended to use a minimum of four cameras, the positions of which will ultimately be dictated by the nature of the data requirements. The red dots (and associated shaded areas indicating the fields of view) in Figure 5.1 show a basic layout for capturing good quality video data about the driver, the vehicle and the environment. It is clear, how- ever, that if an alternative shot is required, another camera will have to be introduced or a compromise will have to be made with the existing arrangement. Camera positioning will also be dictated somewhat by other factors, some of which can be difficult to resolve (such as suitable mounting positions or available space within the vehicle) and others which need careful consideration (such as ethical constraints re- stricting images of passengers). In most cases a compromise between meeting the ideal video requirements and finding a practical solution will need to be made.
Figure 5.1 Video configuration for a left-hand drive passenger car Within PROLOGUE, experiences with manual analysis of continuous video recording proved to be time-consuming, in particular if, as preferred, the video data is analyzed by two persons. When triggering thresholds were employed, there appeared to be a large number of ‘false positives’: an event was automatically identified as relevant, but further visual inspectio...
