Surround View Camera System Based On Embedded GPU

The surround view camera system is an emerging automotive ADAS(Advanced Driver Assistance System)technology,it is developed from the vehicle backup camera system,this traditional technology can only cover a limited area in the rear of the vehicle,and the visually-impaired areas around the vehicle and the front of the vehicle increase the dangers of safe driving.The surround view system is mainly composed of wide-angle cameras,image capture components and image processing components.It processes multi-channel images collected at the same time in real time to form a 360° view around the carrier,eliminating blind spots of vision and providing the viewer of the environment around the carrier in real-time,significantly improves safety.This paper first studies the framework of the surround view system and builds hardware and software platforms around image capture and processing.The hardware mainly includes fish-eye cameras,image capture modules and image processing modules.The software is subdivided into parameter calibration and real-time display.Then study the model of fish-eye camera,focusing on the analysis of the point to camera coordinates in the world coordinates,and then to the image plane coordinates,introduce the linear internal camera parameters to model the linear error.And introduce four common fisheye lens projection models with general nonlinear error internal parameters.And then the Zhang's calibration algorithm is obtained by using OpenCV to obtain two kinds of internal parameters.Second we study the 2D surround view system started with the interpolation and mapping techniques commonly used in image transformation.In order to realize the real-time display of surround view system,we take the static parameters for image stitching,use V4L2 to capture images,and then import them to OpenGL for real-time calculation,the image mosaic seam is eliminated by using the blend channel in OpenGL.Finally,this paper introduces the 3D surround view system to overcome the defects such as limited field of view and stretched edge image in 2D surround view,uses 3ds Max to build a 3D bowl mesh model,and imports it into OpenGL for pseudo 3D reconstruction,using virtual camera to roam the 3D environment,so that each scene in the grid is traversed and the blind spot of the field of view is eliminated.This paper starts from the 2D surround view to the 3D surround view,studies the key algorithm among them,and verifies on the hardware platform.Experiments show that the surround view system based on embedded GPU proposed in this paper has excellent performance.The main calculation is performed by GPU and only occupies a small amount of CPU resources.And using the virtual camera to achieve the perspective of the roaming in the 3D scene,observe every corner around the carrier,to overcome the defects of the 2D surround view.