Changes in video detection performance at signalized intersections under different illumination conditions

The performance of three widely used video detection systems (VDS) was evaluated under five illumination conditions: dawn, sunny morning (long shadows), cloudy noon, dusk, and night. Large data sets were collected in stop bar and advance detection zones of a signalized intersection where VDS were installed side by side. Potential detection errors from each system (false, missed, stuck-on, and dropped calls) were initially identified by comparing VDS and loops using a computer algorithm; then they were visually verified by using videos from the intersection. Results showed significant effects of illumination conditions on the three VDS, which performed best in cloudy noon conditions. In the stop bar zones, false calls increased for all systems mostly at dawn (up to 31%), dusk (up to 51%), and night (up to 35%). Missed calls were affected in only one system and one zone with up to 29% missed vehicles; stuck-on calls increased up to 8% in sunny conditions and up to 6% at night. Dropped calls were rare, and only three were observed in all conditions and for all systems. In the advance zones, the three systems had the most false calls during sunny conditions (up to 53%); missed calls increased for only one system (up to 18%)-also in sunny conditions; and stuck-on calls were less than 2%. This evaluation would help to identify strengths and weaknesses of VDS at signalized intersections, causes of detection errors, and potential effects in traffic operation and safety.