| With the rapid development of high speed and high-resolution digital image technology,digital image correlation(DIC)measurement technology has been well applied in the threedimensional dynamic deformation measurement of structures.However,due to the increase of spatial resolution and the improvement of the acquisition frame rate,the amount of data that needs to be transmitted and stored is also greatly increased.Especially when a multi-camera network is used to measure the dynamic deformation of large structures,the limited storage space often restricts the experiment duration and cannot meet the requirement of destructive experiments.At the same time,with the combination of DIC technology and modern computer technology,distributed high-speed DIC scheme will be the focus of future research.The huge amount of data will bring obstacles to the transmission of each computing node in distributed DIC system.To deploy high-speed camera real-time image acquisition system of high precision dynamic deformation of large space structure,improve the existing data acquisition bandwidth,extend working hours of system and lay the foundation of data transfer in distributed DIC,a GPU parallel accelerate FELICS lossless image compression method is proposed in this paper.A high-speed camera system is designed which can capture,compress and store images in realtime at up to 500 FPS.This paper designs a high-speed DIC camera acquisition,compression,and storage system,which gives full play to the performance of CPU-GPU heterogeneous computing.It mainly consists of four parts: front-end acquisition module,data compression module,storage module,and cache area.At the acquisition end,Coax Press high-speed camera and acquisition card with excellent performance indexes were selected,and the industrial vision standard provided by EVMA was used for secondary development to control the front-end acquisition.The FELICS lossless compression algorithm is selected as the main research object in the data compression module.Based on GPU and CUDA platforms,nearly 4 million threads are used to carry out a pixel-level parallel design.Combined with the parallel prefix algorithm,the parallel packaging of code streams of variable length is completed.The storage module adopts the high-speed SSD with M.2 interface as the storage medium,and the storage mode is set as binary code stream in the software.Finally,a two-level cache area is constructed by using the asynchronous ring queue based on the producer-consumer model to coordinate the asynchronous concurrent work of the first three modules.This paper verifies the whole design scheme through the experiment.Firstly,the simulation performance of the GPU-based parallel FELICS coding kernel is tested.The experimental results show that the parallel compression algorithm has a data compression ratio of nearly 2 times or more for all kinds of speckle images,and the average encoding time of a single 4-megapixel image is less than 2ms.Then,the two systems designed in this paper are used for image acquisition with 100 Hz vibration at 500 FPS.The experimental results show that the system can realize stable acquisition,compression and storage for up to 30 minutes at500 FPS,and the accuracy of DIC calculation is not affected.Under the same experimental time,the system with data compression function only needs half of the original storage space,and the amount of data is greatly reduced,which lays a foundation for the deployment of multicamera systems and distributed transmission. |
PDF Full Text Request