Starling Intelligence at Abbey Road Pedestrian Crossing
In 2020 we ran a Starling Intelligence test using the Abbey Road pedestrian crossing to show that the quality and the amount of the data and detailed information is of more use for real-time assessment or historical analysis.
Goal and Process of the Experiment
This experiment aimed to demonstrate that Starling Intelligence can gather data and information associated with the crossing used as a tool for road user behaviour surveys.
Road user behaviour surveys use recorded video footage that a human analyses. For this test, we used footage captured from a public webcam, then using Starling Intelligence and a human operator to analysis the video, and then compare the results.
Context to the Abbey Road Pedestrian Crossing Test
The crossing was made famous from the cover of The Beatles’ album of the same name. The crossing has many tourists reenacting the cover and taking photos that create dangerous situations for drivers and cyclists.
The area that the experiment covered 40m x 15m around the crossing. The traffic travels along a northbound and sound-bound axis. Whereas the crossing flow is along the eastbound and westbound axis.
Technical Challenge
A camera calibration process is required to get an accurate 2D-3D translation, turning camera coordinates into real-world coordinates. We developed the Camera Auto-Calibration feature that calibrates all the cameras without the need for physical access to the camera to do a manual calibration.
The Metrics and the Results
Following are examples of the results from comparing a human analysis with Starling Intelligence.
Density
This result shows a basic count of total road users on the scene, separated by all vehicles and pedestrians.
The difference between the human analysis and Starling Intelligence is 1.30% for vehicles and 6.45% for pedestrians.
Traffic Flow
This result shows a vehicle count separated by direction of traffic flow, northbound and southbound.
The difference between the human analysis and Starling Intelligence is 1.30% for total traffic, 6.06% for southbound traffic, and 6.82% for southbound traffic.
Crosser Flow
This result shows a crossers count separated by direction of moving eastbound and westbound.
The difference between the human analysis and Starling Intelligence is 0 for total crossers, 9.09% for eastbound crossers, and 9.09% for westbound crossers.
Vehicles that Give Way
This result shows counts vehicles that either give way to allow pedestrians to cross or not.
The difference between the human analysis and Starling Intelligence is 9.09% for vehicles that give way and 0% for vehicles that do not give way.
Crosser Behaviours
These result separates the crossers by their behaviour:
Confident crosser: crossers that cross without modifying speed, the difference is 0%
Reluctant crosser: crossers that wait for the vehicle to stop before crossing, the difference is 60%
Running crosser: crossers that ran while crossing, the difference is 0%
Denied crosser: crossers that can not cross because a vehicle did not stop, the difference is 0%
Almost hit: crossers that are almost got hit by the vehicle while crossing, the difference is 0%
Irrelevant crosser: crossers that cross without any vehicles present, the difference is 0%.
Vehicle Behaviours
These results count individual vehicles by their behaviour:
Vehicles that are slow enough to let pedestrians confidently cross, the difference is 0%
Vehicles that are too fast but stopped to let pedestrians cross, the difference is 60%
Vehicles that are too fast and force pedestrians to run across the crossing, the difference is 0%
Vehicles that do not stop at all, the difference is 0%
Vehicles that almost hit pedestrians, the difference is 0%