Research On Lane Detection Algorithm Based On CUDA Acceleration

In recent years,with the rapid development of information technology,artificial intelligence and other fields,the traditional automobile manufacturing industry has been transformed.Intellectualization has become an inevitable trend in the development of automobile industry.The key of vehicle intelligentization lies in perception,decision and control,and the technology of vehicle environment perception is the basis of vehicle path planning and decision-making.Lane detection technology,as an important part of vehicle environment perception,has become the basis of many advanced assistant driving systems.Therefore,this paper studies the lane detection algorithm based on CUDA acceleration.First of all,this paper develops a lane detection algorithm with good environmental adaptability for the complex urban road environment.The process consists of five steps: image preprocessing,Canny edge detection,Hough transform line detection,Lane candidate selection and curve fitting.Secondly,in order to improve the real-time performance of lane detection,combined with the latest development of GPU in the field of image processing,the CUDA Programming Model and storage model based on GPU are described,and the CUDA parallel optimization strategy is analyzed as the basis for the subsequent parallel optimization of lane detection algorithm.Then,the algorithm of lane detection is optimized based on CUDA,and the image is preprocessed by using the CUDA module library of OpenCV.Aiming at the classic Hough transform line detection method,an improved parallel Hough transform detection method is proposed,which uses a large number of threads to independently calculate the spatial coordinate conversion instead of the serial cycle calculation.For lane selection,CUDA parallelism is also used to make each line Each procedure corresponds to a set of parameters to judge the constraints.Finally,collect the urban road scene as the experimental data set,based on the server and embedded development board TX2 as the experimental platform,carry out the lane detection experiment under the complex traffic scene.Compared with CUDA acceleration optimization before and after lane detection effect,and time-consuming statistics and analysis,verify the effectiveness and real-time of the improved parallel lane detection algorithm.