Lane Detection Using Segmented Neural Network Analysis

The patent segments the image processing task into two distinct stages: a first stage using traditional edge detection methods to identify candidate regions, and a second stage using neural networks to classify only those candidate regions. This segmentation allows the computationally intensive neural network to process only a small subset of image pixels (those in candidate regions), rather than all pixels, thereby maintaining detection accuracy while significantly reducing energy consumption.

The patent applies preliminary action by performing traditional edge detection and candidate region identification before applying the neural network. This pre-processing step filters out large portions of the image that do not contain lane markings, so that the neural network only needs to process pre-identified candidate regions. This preliminary filtering action reduces the computational load and energy requirements of the subsequent neural network classification.

The patent divides the image evaluation process into two segmented stages: rapid traditional edge detection for initial candidate identification, followed by neural network classification only for those candidates. This segmentation enables the system to maintain high evaluation speed by avoiding neural network processing of entire images, while still achieving high detection accuracy through targeted neural network analysis of candidate regions.

The patent performs preliminary edge detection and candidate region identification using computationally efficient traditional methods before applying the slower but more accurate neural network. This preliminary action filters the image content so that the neural network processes only relevant candidate regions, thereby maintaining high evaluation speed while improving detection accuracy in the final classification stage.

The patent segments the detection task into two functional parts: traditional edge detection for efficient candidate region identification, and neural network classification for accurate lane marking verification. This segmentation allows the system to leverage the low computational cost of traditional methods for initial filtering, while using the high accuracy of neural networks for final classification, thereby achieving both low overall computational cost and high detection accuracy.

The patent merges two different detection approaches into a unified two-stage system: traditional edge detection methods and neural network classification. By combining these methods in sequence rather than choosing one or the other, the system achieves the computational efficiency of traditional methods for initial processing and the detection accuracy of neural networks for final classification, thereby resolving the contradiction between computational cost and detection accuracy.

The patent segments the lane detection process into two stages with different speed-accuracy characteristics: a fast traditional edge detection stage for rapid candidate identification, and a more accurate but slower neural network stage for precise classification. This segmentation enables the system to maintain high overall processing speed by limiting neural network processing to only candidate regions, while achieving high detection accuracy through the neural network's precise classification of those candidates.

The patent merges traditional edge detection methods and neural network classification into a hybrid two-stage system. The traditional methods provide rapid initial processing and candidate generation, while the neural network provides accurate final classification. By merging these approaches in sequence, the system achieves both high processing speed from the traditional methods and high detection accuracy from the neural network, resolving the contradiction between speed and accuracy.