Camera-Based Tolling Using ML for Axle Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current toll systems rely on expensive and complex sensors embedded in roadways, which are costly to install and maintain, and require manual review of low-confidence license plate reads, leading to high operational costs and inefficiencies.
Innovation Solution
A system utilizing machine learning models and cameras for vehicle detection, classification, and tracking, which uses convolutional neural networks to identify vehicles and recognize license plates without the need for embedded sensors, employing passive optical systems and computing resources to reduce installation and maintenance costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If axle sensors are embedded in roadways for vehicle detection, then vehicle identification accuracy is improved, but installation cost and maintenance complexity increase
Solution Approach 1:
The patent replaces the mechanical/physical axle sensors embedded in roadways with an optical detection system using cameras and machine learning algorithms. The camera captures images of vehicles, and the ML model processes these images to identify vehicle characteristics (number of axles, vehicle type) without requiring physical contact or embedding sensors in the roadway infrastructure.
Solution Approach 2:
Instead of using physical sensors that directly detect vehicle axles, the system creates optical copies (images) of vehicles and analyzes these copies through machine learning models. The ML model processes visual data to extract information about vehicle axles and characteristics, replacing the need for direct physical measurement.
2Reliability
If axle sensors are embedded in roadways for vehicle detection, then toll collection reliability is improved, but maintenance cost increases
Solution Approach 1:
The patent replaces embedded axle sensors with a camera-based optical detection system. Cameras are mounted on existing infrastructure (such as toll booth structures) rather than being embedded in the roadway, making them easier to access for maintenance and repair while maintaining reliable vehicle detection and toll collection capabilities.
3Measurement precision
If manual review of low-confidence license plate reads is implemented, then identification accuracy is improved, but operational time and cost increase
Solution Approach 1:
The system implements self-service through automated machine learning models that independently process and verify license plate readings. The ML model continuously learns from data and automatically adjusts its recognition thresholds, eliminating the need for manual review of low-confidence reads while maintaining high identification accuracy.
Solution Approach 2:
The patent incorporates feedback mechanisms where the system automatically reviews and corrects its own license plate readings. The ML model uses feedback from verified readings to continuously improve its accuracy, replacing manual human review with automated iterative improvement that reduces operational time while maintaining precision.
Data Source
AI summary
Generally discussed herein are systems, devices, and methods for improved automatic toll generation. A system can include one or more cameras positioned to capture images of vehicles passing through a toll station, vehicle detection and classification circuitry to receive an image of a vehicle from the one or more cameras and execute a first machine learning (ML) model to classify the vehicle, axle count circuitry to, in response to receiving data from the vehicle detection and classification circuitry that a vehicle was detected, execute a second ML model to classify a number of axles of the vehicle for determining a toll charge, and toll circuitry to issue the toll charge based on the classified number of axles.