The detection performance of video detection systems at intersections, from three major manufacturers, was evaluated using a side-by-side installation and large datasets. The performance at three stop bar detection zones under two illumination conditions is presented: day with no shadows (cloudy days) and night. In addition, these two conditions and day time with shadows (sunny mornings) are also evaluated under windy conditions. Four measures of performance were used to quantify the errors in the detection: false calls, missed calls, stuck-on calls, and dropped calls. Activation and deactivation times of the detection zones were used to automate the initial calculation of errors, while video images from the intersection were used for their manual verification. Night time results showed an average decrease in false calls for the left-most lane (from about 14% to 9%), but an increase for the middle lane (from 1.5% to 23%) and thru-right lane (from 0.8% to 13%). Vehicle headlights reflecting on adjacent zones and the viewangle of the cameras were important factors in false calls. Missed calls in night time increased only for one system in one lane (from 0.2% to 26%). Wind increased false calls, mostly in sunny morning conditions (long shadows) between 12% and 17% on average, and reduced stuck-on calls due to the camera oscillation movement. Missed calls increased in day time only, from 0.9% to 2.4%, but it was not possible to identify the exact cause for this increase after a visual inspection of the videos.
Abstract