LEARNING METHOD LEARNING DEVICE FOR DETECTING LANE THROUGH CLASSIFICATION OF LANE CANDIDATE PIXELS AND TESTING METHOD TESTING DEVICE USING THE SAME

摘要

The present invention provides a learning method for detecting at least one lane based on a Convolutional Neural Network (CNN), comprising: (a) obtaining, by a learning apparatus, information on an encoded feature map and a lane candidate pixel in an input image; (b) the learning device classifies a first part of the lane candidate pixel having a probability score less than a preset threshold as a strong line pixel, and classifying the second part of the lane candidate pixel having a probability score less than the threshold value and higher than the other threshold value as a weak line pixel classifying as; and (c) when the distance between the weak line pixel and the strong line pixel is less than the preset distance, the learning device classifies the weak line pixel as a pixel corresponding to the additional strong line, and the pixel corresponding to the strong line and the additional strong line The method comprising the step of determining that the pixel corresponds to the lane corresponding to the lane;