Abstract: A dual mode method of assessing a toll fee by collecting i) sensor data of a vehicle as it travels on a lane and determining a first axle count based on said sensor data; and 2) collecting video images of said vehicle and determining a second axle count based on said video images. The video images are analyzed using computer vision and machine learning to determine the second axle count, and the first and second axles count compared and if identical the vehicle is charged based on the axle count. However, the video images are re-analyzed if the first and second axle counts are not identical and this is repeated until identity is obtained. The system learns from its errors, and the axle count improves with time. If identity is not reached, the transaction record can be flagged for human intervention, but tests show at least 100 fold reduction in errors, thus obviating thousands of interventions per day. Ultimately, monthly summaries of charges are compiled and sent to the customer.