Real Time Measurement Of Background Oriented Schlieren Based On GPU

Background oriented schlieren measurement technology is widely used in aerospace,aerodynamics,and other fields.It not only can measure the flow field without contact and low cost but also can be extended to the flame temperature field,the gas refractive index field,and other research,which is of great significance to production and life.But in the current era of big data,camera data collection is increasing too,what's more,the real-time requirements for background oriented schlieren measurement technology are increasing.Therefore,this paper focuses on the real-time measurement technology of background shadow.At the first time,his paper compares the advantages and disadvantages of different measurement methods from the point displacement of a background image,which is the key of background image measurement,and analyzes the theory of background-oriented schlieren measurement technology based on eight neighborhood feature point recognition.The method of local adaptive threshold is proposed to calculate the feature area,which improves the effect of preprocessing,and the anti-interference of feature recognition is improved.For a second,the hardware structure of GPU and the principle of CUDA Programming are studied.Combined with the principle of BOS measurement based on feature point recognition,the parallel analysis is carried out by sub-modules.The parallel acceleration method of BOS based on CPU-GPU is realized: image preprocessing module uses shared memory instead of global memory to access and improves the computing speed;coordinate and displacement calculation adopts the method of protocol sum,reducing Low algorithm complexity;The method of row and column recognition is designed,which has higher parallelism,faster processing speed,and stronger real-time performance.In addition,according to the different ways of thread allocation,the paper analyzes the influence of thread allocation on the running time of the algorithm,establishes the time consumption model of each module,studies the factors affecting the resource consumption of each module,which plays an important reference value for the design of experimental conditions.Finally,the system is designed and built based on CPU-GPU.The CPU realizes the functions of parameter setting,reference image data processing,image and solution results display,GPU end realizes the algorithm of feature point recognition and displacement vector calculation.The whole system can realize the BOS measurement technology with row and column and feature point recognition as the core,and can achieve real-time processing of3.46G/s at most Speed.Finally,the real-time and feasibility of the system are verified by experiments.