Development Of 4K Endoscope Image Processing Algorithm Based On Tegra X1 Processor

As an important clinical diagnostic and therapeutic equipment,the medical electronic endoscope has a high demand on the image quality.With the advent of 4K display technology,the image resolution of endoscopy should also be improved.Because 4K resolution image has 4 times the data size of 1080P image,it is necessary to develop a more efficient image processing algorithm to ensure real-time and low-latency image display of electronic endoscope.This thesis develops a 4K endoscope image processing algorithm using the GPU parallel processing technology on the hardware platform with two Tegra X1 processors.The main part of the algorithm is real-time image front-end processing,which is used to process the raw data acquired by endoscope for displaying.The other part of the algorithm is non-real-time H.264 video encoding,which utilizes the residual performance of GPU to encode and store important data.According to the CUD A programming model and the specific hardware configuration of Tegra X1,this thesis designs the thread allocation scheme separately for each module to improve the total parallel efficiency.In the algorithm optimization,this thesis uses shared memory to reduce the read and write times of global memory and adjusts the thread processing order to decrease the program conditions branch.In addition,this thesis also uses zero copy technology to reduce the CPU-GPU data communication time.In the algorithm co-operation implement,this thesis uses the stream processing to eliminate the interference of different processes using the same GPU resources,and further divide the modules in the encoding algorithm,thus shortening the continuous occupation time of the GPU to ensure the real-time of image front-end processing.The experimental results show that the algorithm developed in this thesis can guarantee the 30fps real-time image front-end processing while encoding 4K video at the speed of 0.7 to 1.0fps.When the video encoding algorithm is not running,the speed of image front-end processing can reach 56fps.In addition,this thesis evaluates the parallel optimization effect of each module of H.264 video encoder,which is expected to provide reference for other researches about video encoding on GPU.