CNN Lane Detection via Strong Weak Pixel Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional lane detection methods using convolutional neural networks (CNNs) struggle to accurately detect lanes, especially those at a distance or with low probability pixels, leading to the generation of disconnected lines and difficulty in identifying long lines.
Innovation Solution
The proposed method involves a learning device that applies convolution and deconvolution operations to input images to classify lane candidate pixels based on probability scores and distance criteria, using strong and weak line classifications to refine lane detection, and employs Polynomial or spline fitting algorithms to determine lane pixels, optimizing CNN parameters through segmentation loss calculation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional CNN segmentation is used to detect lanes, then the detection process is simple, but lanes at distance and with low probability pixels are not accurately detected, leading to disconnected lines
Solution Approach 1:
The patent segments the lane detection process into multiple stages: initial CNN segmentation to obtain candidate pixels, probability thresholding to classify strong/weak line pixels, and iterative line extension to connect segments. This multi-stage segmentation approach improves detection accuracy by handling different pixel confidence levels separately while maintaining manageable complexity through modular processing.
Solution Approach 2:
The patent performs preliminary classification of candidate pixels into strong line pixels (high probability) and weak line pixels (low probability) before actual lane connection. This preliminary action allows the system to prioritize high-confidence pixels while systematically incorporating low-confidence pixels, improving overall detection accuracy without overwhelming complexity.
2Productivity
If only high probability pixels are used for lane detection, then the detection process is fast, but many disconnected lines are generated and long lines cannot be obtained
Solution Approach 1:
The patent performs preliminary classification of candidate pixels into strong line pixels (high probability) and weak line pixels (low probability) before actual lane connection. This preliminary action allows the system to prioritize high-confidence pixels for fast initial detection while systematically incorporating low-confidence pixels to ensure lane continuity and reduce disconnected lines.
Solution Approach 2:
The patent implements a dynamic, iterative process where lines are progressively extended by incorporating weak line pixels adjacent to existing strong lines. The detection process adapts by continuously evaluating probability thresholds and expanding line segments, balancing speed and reliability through dynamic adjustment of detection criteria.
3Measurement precision
If low probability pixels are excluded from lane detection, then the detection accuracy for clear lanes is high, but thin lanes and distant lanes are missed
Solution Approach 1:
The patent applies different quality criteria to different regions and pixel types: strong line pixels (high probability) form the core of detected lanes with high confidence, while weak line pixels (low probability) are selectively incorporated to extend lines and detect thin or distant lanes. This local quality differentiation allows the system to maintain high accuracy for clear lanes while expanding detection coverage for challenging cases.
Solution Approach 2:
The patent implements a dynamic, iterative process where lines are progressively extended by incorporating weak line pixels adjacent to existing strong lines. The detection process adapts by continuously evaluating probability thresholds and expanding line segments, balancing speed and reliability through dynamic adjustment of detection criteria.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A learning method for detecting at least one lane based on a convolutional neural network (CNN) is provided. The learning method includes steps of: (a) a learning device obtaining encoded feature maps, and information on lane candidate pixels in a input image; (b) the learning device, classifying a first parts of the lane candidate pixels ,whose probability scores are not smaller than a predetermined threshold, as strong line pixels, and classifying the second parts of the lane candidate pixels, whose probability scores are less than the threshold but not less than another predetermined threshold, as weak lines pixels; and (c) the learning device, if distances between the weak line pixels and the strong line pixels are less than a predetermined distance, classifying the weak line pixels as pixels of additional strong lines, and determining that the pixels of the strong line and the additional correspond to pixels of the lane.