Brief description

In recent years, the introduction of augmented reality in ADAS has become an active research topic. Such a system contributes to a more controllable, comfortable, and safe driving in the city. The surround view is one such system, which could be expanded by artificial intelligence. This paper presents one approach to designing a real-time surround view multi-camera system on Jetson Tegra TX2. The algorithm for finding intrinsic and extrinsic camera parameters based on compute homography matrix from descriptors of feature points was implemented on CPU and computing only once. To achieve real-time, the algorithms of geometric, photometric alignment and blending were implemented on GPU and use CUDA technology. Texture mapping on 3D bowl mesh using OpenGL ES. A virtual camera is used to achieve a 360° view. Experiment results show a surround view system with a good stitched quality and acceptable performance. Further development of the research project is presented.