Panorama Based Lane-level HD-Map Generation

As the development of autonomous driving and advanced driver assistance system,the road data become more and more important.It's difficult for the sensor system on vehicles to detect and recognize the environment when it comes to the complex urban road.Thus,the researchers begin to use road data as prior knowledge to support other sensors in completing their missions.Normal road maps only contain limited species road data with low accuracy,however,what the intelligence driving systems need is HD-Map.The HD-Map is a kind of digital map which provides almost all kinds of accurate and detailed road data that is useful for intelligence driving systems,a HD-Map generation method based on panorama is proposed in this paper.Firstly,taking the function of HD-Map into consideration,a data collecting and processioning system with panorama camera as core was designed,the panorama camera was used to collect road images,the positioning data acquisition sensors included GPS,inertial navigation system and wheel speed encoder.Got the vertical images and related gesture information with the data collected,then,mosaicked the high resolution and accurate map on with could we label the road data.The road data labeled on the map contained its position and properties information.Secondly,the HD-Map should be optimized to improve its accuracy and visual effect.In order to get the correct vertical images,gnomonic projection was adopted,by detecting the parallel state between traffic lanes and road edges,the error of pitch in gnomonic projection could be measured and corrected.In order to remove the obstacles in vertical images,LIDARs were used to detect the position of obstacles in verticals images,then the images were sliced and a map without obstacles was re-mosaicked.Lastly,since only two-dimensional information could be got from the map,gnomonic projection was applied to get images of the road in front,with which a binocular vision system could be simulated between two frames,the height of traffic facilities could be calculated with this system.The HD-Map generated with the method in this paper contains detailed road data whose accuracy can reach sub-meter level.