Autonomous vehicles aim at higher levels of intelligence to recognize all the elements in the surrounding environment; in order to be able to make decisions efficiently and in real time. For this reason, a convolutional neural networks capable of perform semantic segmentation of these elements have been implemented. In this work it is proposed to use the ERFNet architecture to segment the main obstacles and lanes in a road environment. One of the requirements for training this type of networks is to have a complete and large dataset with these two types of labels. In order to avoid manual labeling, an automatic way of carrying out this process is proposed, using convolutional neural networks and different dataset already labeled. The generated dataset contains 19000 images tagged with obstacles and lanes, to be used to train a network of ERFnet architecture. From the experiment, the obtained results show the performance of the proposed approach providing accuracy of 74.42%.
Joint Instance Segmentation of Obstacles and Lanes Using Convolutional Neural Networks
Advs in Intelligent Syst., Computing
Iberian Robotics conference ; 2019 ; Porto, Portugal November 20, 2019 - November 22, 2019
2019-11-20
13 pages
Article/Chapter (Book)
Electronic Resource
English
The detection of lanes and obstacles in real time using optimal moving window
Tema Archive | 2001
|CLARK: a heterogeneous sensor fusion method for finding lanes and obstacles
British Library Online Contents | 2000
|