CNN Lane Detection Clustering Reduces Computational Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional lane detection methods using CNNs require extensive computation as they need to check every pixel in the input image to determine whether it belongs to a lane, making them inefficient.
Innovation Solution
The method clusters lane pixels into groups, calculates representative values for each cluster, and uses existence and distance regression losses to optimize CNN parameters, reducing the need to check every pixel by focusing on feature maps near estimated lanes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional lane detection methods check every pixel in the input image to determine lane membership, then detection accuracy is maintained, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the image processing task by first detecting candidate lane regions using a CNN, then performing detailed lane pixel identification only within these segmented regions rather than checking every pixel in the entire image. This segmentation approach maintains detection accuracy while significantly reducing computational complexity by limiting the detailed analysis to relevant areas only.
Solution Approach 2:
The patent applies preliminary action by using a CNN to pre-identify candidate lane regions and generate probability maps before performing the actual lane detection. This preliminary processing step filters out non-lane areas, allowing the subsequent detection algorithm to focus computational resources only on regions with high lane probability, thereby reducing overall computational complexity while maintaining accuracy.
2Loss of information
If conventional methods process every pixel to determine lane membership, then complete lane information is obtained, but processing time increases
Solution Approach 1:
The patent segments the processing domain into candidate lane regions identified by the CNN and non-lane regions. By applying detailed processing only to the segmented candidate regions and using probability thresholds to filter results, the method recovers complete lane information while reducing processing time through selective processing of only relevant image portions.
Solution Approach 2:
The patent applies partial action by processing only the subset of pixels within candidate lane regions rather than all pixels in the image. The CNN-generated probability maps guide this partial processing, allowing the system to obtain complete lane information by focusing computational effort on areas with high lane probability while ignoring background areas.
3Measurement precision
If the CNN processes the entire image to detect lanes, then all lane pixels are identified, but computational resources are wasted on non-lane areas
Solution Approach 1:
The patent extracts candidate lane regions from the entire image using the CNN's probability mapping capability. By extracting and isolating only the regions with high lane probability, the method enables subsequent processing to focus exclusively on these extracted regions, thereby maintaining lane pixel identification accuracy while significantly reducing computational energy consumption by excluding non-lane areas from detailed processing.
Solution Approach 2:
The patent applies local quality by assigning different processing intensities to different regions of the image based on CNN-generated probability values. High-probability regions undergo detailed lane pixel identification while low-probability regions are either skipped or processed with reduced intensity, optimizing the balance between detection accuracy and computational energy consumption through spatially varying processing quality.
Data Source
Figure 1(A)~1(C)
Figure 2
Figure 3
AI summary
A learning method of a CNN for detecting lanes is provided. The method includes steps of: a learning device (a) instructing convolutional layers to generate feature maps by applying convolution operations to an input image from an image data set; (b) instructing an FC layer to generate an estimated result vector of cluster ID classifications of the lanes by feeding a specific feature map among the feature maps into the FC layer; and (c) instructing a loss layer to generate a classification loss by referring to the estimated result vector and a cluster ID GT vector, and backpropagate the classification loss, to optimize device parameters of the CNN; wherein the cluster ID GT vector is GT information on probabilities of being cluster IDs per each of cluster groups assigned to function parameters of a lane modeling function by clustering the function parameters based on information on the lanes.