Introductioin

This is a research project supported by HERE company. Therefore, the datasets(which will be published very soon) and some codes of the algorithm are confidential so won't be shown in this repo(the .If you want to learn more about the dataset and algorithm, please read this project's paper Lane Boundary Extraction from Satellite Imagery, which is accepted by ACM SIGSPATIAL Workshop 2017. However, we will provide some useful codes and tools such as satellite map retriving from Bing Map，diffrent coordinates transforamtion， evluation algorithm and a high-level easy-plug-in deep learning toolbox to deal with images written by tensorflow

Automated driving is becoming a reality. In this new reality, High Definition (HD) Maps play an important role in path planning and vehicle localization. Lane boundary geometry is one of the key components of an HD Map. Such maps are typically created from ground level LiDAR and imagery data, while useful in many ways, have many limitations such as prohibitive cost, infrequent update, traffic occlusions, and incomplete coverage. In this project, we propose a novel method to automatically extract lane boundary from satellite imagery using pixel-wise segmentation and machine learning, and convert unstructured lines into structured road model by using hypothesis linking algorithm, which addresses the aforementioned limitations. We will also publish our experiment dataset consisting of satellite imagery and the corresponding lane boundaries as ground truth to train, test, and evaluate algorithms in the future.