Road Type Classification Using Convolutional Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the type of road on which an autonomous or semi-autonomous vehicle is driving, such as GPS and sensor-based systems, are unreliable due to positioning errors and environmental factors like fog, which can lead to incorrect activation of autonomous driving functions.
Innovation Solution
A method using a deep convolutional neural network that learns to analyze images of the road environment to differentiate between various road types, eliminating the need for predefined characteristic elements and improving robustness by processing the entire driving scene.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If GPS positioning is used to determine road type, then the system is simple to implement, but the reliability is insufficient due to positioning errors
Solution Approach 1:
The patent introduces an intermediary system (sensor-based road type confirmation system) that mediates between GPS positioning and autonomous driving activation. This intermediary uses cameras and other sensors to detect road characteristics and verify whether the vehicle is actually on a suitable road, thereby resolving the reliability issue without completely replacing the simple GPS system.
Solution Approach 2:
The patent replaces the purely positioning-based mechanical/GPS system with a sensor-based detection system that uses cameras and image processing to determine road type. This substitution uses optical detection instead of relying solely on positional data, improving reliability while maintaining reasonable system complexity.
2Reliability
If sensor-based road type confirmation is added, then the reliability improves, but the device complexity increases
Solution Approach 1:
The patent makes the sensor system multi-functional by using the same cameras and sensors for both road type determination and other autonomous driving functions. This universality allows the system to improve road type detection reliability while avoiding the complexity increase that would result from adding dedicated separate sensors.
Solution Approach 2:
The patent merges the road type confirmation function with the existing sensor suite used for autonomous driving. Instead of adding separate dedicated sensors, the system combines road type detection with the existing camera and sensor array, thereby improving reliability without proportionally increasing device complexity.
3Measurement precision
If specific road elements are detected, then the measurement precision is high, but the reliability deteriorates in adverse conditions like fog or obscured signs
Solution Approach 1:
The patent segments the road type determination process into multiple independent detection channels (road markings, signs, vegetation patterns, terrain features). Instead of relying on a single specific road element, the system analyzes multiple segments of visual information, so that if one element is obscured by fog or other vehicles, other segments can still provide reliable classification.
Solution Approach 2:
The patent changes the detection parameters from specific, discrete road elements (signs, markings) to more general scene characteristics (vegetation patterns, terrain slope, overall road geometry). This parameter change makes the detection less sensitive to adverse conditions that obscure specific elements, thereby improving reliability while maintaining measurement precision through pattern recognition.
Data Source
Figure 1
Figure 2
AI summary
A method for determining the type of road on which a vehicle is travelling based on a plurality of acquired images of the road, said method comprising a first, preliminary phase (10) comprising the steps of preprocessing (200, 300) the images and learning (400) a neural network (RN) based on pairs (image, type of road); said method comprising a second, implementation (500) phase (20), in which the coefficients optimised in the learning step (400) are entered into a computer (ECU) of the vehicle (V) implementing said neural network (RN) and which provides an estimate of the type of road on which the vehicle (V) is travelling based on the images acquired by at least one image capturing device (CAM) of the vehicle (V) and the optimised coefficients.