Method and device for generating image data set to be used for learning CNN (Convolutional Neural Network)

Abstract

A method of generating at least one training data set includes steps of: (a) a computing device acquiring (i) an original image and (ii) an initial synthesized label generated by using an original label and a bounding box corresponding to an arbitrary specific object; and (b) the computing device supporting a CNN module to generate a first synthesized image and a first synthesized label by using the original image and the initial synthesized label, wherein the first synthesized label is created by adding a specific label to the original label at a location in the original label corresponding to a location of the bounding box in the initial synthesized label, and the first synthesized image is created by adding a specific image to the original image at a location in the original image corresponding to the location of the bounding box in the initial synthesized label.

Description

technical field [0001] The invention relates to a method and a computing device for generating at least one CNN learning image dataset for detecting at least one obstacle in an automated driving situation. Background technique [0002] Deep Convolutional Neural Networks (Deep CNNs) are at the heart of amazing developments in the field of deep learning. CNNs have been used to solve text recognition problems in the 90s, but their widespread use today is due to recent research results. The above-mentioned deep CNN won the championship in the ImageNet image classification competition in 2012. Convolutional neural networks have since become very useful tools in the field of Machine Learning. [0003] In addition, image segmentation (Image segmentation) is a method of receiving an image (training image or test image) as input and creating a label (label) as output. Recently, with the attention of deep learning (Deep learning) technology, image segmentation tends to use more dee...

AI Technical Summary

Problems solved by technology

That is, it is difficult to obtain learning images for special objects that do not often appear on the road from general driving image data

For example, it is easier to obtain image data for people, bicycles, and vehicles from general driving image data, so in order to improve detection performance, learning can be performed on images containing these objects, but it is difficult to obtain images for tigers or crocodiles from general driving image data data, so there is a problem that it is not easy to learn to improve the detection performance of special objects

Images