Road Orthophoto Map Generation Based On Multiple Onboard Cameras

In recent years, maps play an increasingly important role in aided driving. In the future, aided driving will become more and more intelligent, and researchers will probably realize real intelligent driving finally. In intelligent driving, vehicles need to get rich road surface information to realize autonomous positioning and navigation, such as lane markings, stop lines, road surface traffic signs, etc. However, since the resolution of traditional satellite or aerial maps is not high enough, such maps cannot clearly reflect detail information of roads, In this case, a lane-level road orthophoto map generation method using multiple onboard cameras is proposed in this thesis. Through this method, a road orthophoto map with high resolution can be generated. The detailed research contents of this thesis are as follows:Firstly, towards image data capture, a multiple onboard camera capture system is constructed in this thesis. Compared with the capture system using the panoramic camera, this capture system more applies to road orthophoto map generation of near fields or narrow roads. Besides, the cost of this capture system is much lower than that of the capture system using the panoramic camera; Compared with the capture system using the monocular camera or around view vision system, this capture system has a larger visual field and longer effective visual distance, which means a higher capture efficiency. Towards generation of orthophoto images, wide visual angle cameras, which are used in this thesis, are firstly calibrated to get their intrinsic parameters. The distortion correction of these cameras can be realized with the intrinsic parameters. After that, a joint calibration of these cameras is conducted to realize the generation of orthophoto images. In addition, in order to avoid the appearance of pixel cavities, repeated pixels or color difference, image interpolation and white balance are introduced to optimize image data.Secondly, towards local road orthophoto map generation, a vehicle positioning technology based on the fusion result of differential GPS(Global Positioning System) and INS(Inertial Navigation System) is utilized in this thesis. Through experiment results, we can find out that this positioning method can meet the demand of local road orthophoto map generation; Towards the appearance of overlapped areas, an image fusion method based on the weighted median filter is utilized to fuse pixels in the overlapped areas. Compared with other weighted filter methods, the weighted median filter can help avoid the influence of noise points to finally generated road orthophoto maps. Besides, weighted median filter can ensure road orthophoto maps keep more detailed information and not generate image obscurity and incorrectness.Thirdly, towards large scale road orthophoto map generation, in order to efficiently generate, save and release large scale road orthophoto maps, a scheme based on tile maps is introduced. This scheme constructs a tile pyramid model and introduces the linear quad tree tile index to describe the tile pyramid model. Tile maps with different resolutions are presented in different layers of the tile pyramid, which can avoid wasting a lot of time and memories in large scale image data processing. Besides, considering the compatibility of maps, the map frame based on OSM is introduced during the process of road orthophoto map construction.