Laser Scan Vehicle Classification Using CNN and 3D Point Clouds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle tolling systems face challenges in accurately classifying vehicles due to variations in vehicle design over time, leading to high error rates, especially when environmental factors like precipitation affect camera-based systems, and axle sensors have finite lifespans and are expensive.
Innovation Solution
A vehicle classification system utilizing a laser scan device angled downward to create 3D vehicle images, coupled with a roadside collection unit and a convolutional neural network (CNN) for classification, which includes an informed pseudolabel algorithm to detect labeling errors and an autoencoder for robust vehicle classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If camera-based image analysis is used for vehicle classification, then the system can identify vehicle presence and determine toll charges, but environmental factors such as precipitation adversely affect operation and increase error rates
Solution Approach 1:
The patent replaces camera-based optical detection with laser range finder-based detection. The laser range finder measures distances to vehicle components (bumper, roof, trunk) using time-of-flight measurement of laser pulses, creating 3D point cloud data that is immune to precipitation and lighting conditions, thereby resolving the vulnerability to environmental factors while maintaining classification accuracy
Solution Approach 2:
The patent changes the measurement parameter from 2D image intensity (camera) to 3D spatial distance (laser range finder). By measuring actual distances to vehicle surfaces and constructing point cloud representations, the system captures geometric features that remain consistent regardless of weather conditions, eliminating the harmful effect of precipitation on detection reliability
2Reliability
If axle sensors are used for vehicle detection and classification, then the system can count vehicles and determine toll charges, but the sensors have finite lifespans and are prohibitively expensive
Solution Approach 1:
The patent replaces embedded mechanical axle sensors with non-contact laser range finders mounted on overhead structures. The laser-based system has no moving parts or consumable components, eliminating the finite lifespan issue of axle sensors while maintaining accurate vehicle detection and classification capabilities for toll charging
Solution Approach 2:
The patent creates a virtual 3D model (point cloud) of the vehicle as a copy of its physical form. This digital representation captures all necessary geometric features for classification without requiring physical contact with the vehicle, enabling unlimited detection cycles without wear or degradation of the detection system
3Productivity
If traditional vehicle classification systems are used, then the system can process vehicles, but high error rates occur due to variations in vehicle design over time
Solution Approach 1:
The patent transitions from 2D image analysis to 3D point cloud analysis by adding the depth dimension through laser range finding. This third dimension provides accurate spatial relationships and geometric features that capture vehicle design variations, enabling precise classification across different vehicle types and eras while maintaining high processing throughput
Solution Approach 2:
The patent implements a dynamic classification system using neural networks that adapt to varying vehicle designs. The system processes point cloud data through trainable models that learn to recognize patterns across different vehicle configurations, maintaining high accuracy despite evolutionary changes in vehicle design while preserving processing efficiency
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The system achieves accurate vehicle classification with reduced error rates, even in varying environmental conditions, and extends the lifespan of tolling infrastructure by using a more durable and cost-effective method compared to traditional systems.
Implementation Method 1
A laser scan device angled downward toward incoming traffic coupled to a roadside collection unit (RCU)
Data Source
AI summary
Systems, devices, methods, and computer-readable media for. A method can include receiving, from a laser scan device of a tolling station, a time series of distance measurements, determining, based on the time series of distance measurements, height measurements indicating a height of a vehicle from a surface of a road. generating, based on the height measurements, an image of the height measurements, and classifying, using the image as input to a convolutional neural network (CNN), the vehicle.