A Measuring Algorithm For An Object Based On Multi-view Images And Its Parallel Optimization On GPUs

Because of the low accuracy, long time taken, the large difficult and hardlyautomation, the traditional measurement method has been unable to meet the growingdemands for a variety of measurement. So people pay more attention to the imagemeasurement which has the unique advantages. However, the measurement technologybased on multi-view will bring the increasing demands of the processing data whichmakes the application of image-based measurement facing great challenge of measuringspeed. Therefore, in order to achieve real-time requirements, there is an urgent need toresearch a new approach on how to optimization the procedures of image measurement.In response to the above problems, this paper proposes a measuring algorithm for anobject based on multi-view images and its parallel optimization on GPUs. The multi-viewimage process completes with five sub-processes named the image capture, cameracalibration, stereo matching,3D reconstruction and image measurement. In order to get abetter implementation of the measurement algorithm based on the multi-view images, wetake a stereo matching algorithm based on SIFT and SURF features, and eliminating thefalse match points through the difference comparison between the re-imaging points afterrebuilding for matching point and the original image points. In the reconstruction processof the multi-view, first，we conduct the pairwise triangulation reconstruction, and then getthe exact coordinates reconstruction by obtaining coordinates of the center of gravity.When measuring, the paper select model Temple’s height as a measurement target, thendesigns a efficient measurement algorithm. Experimental result shows that theimage-based measurement approach of the multi-view performs well in calculating theheight of target object, and the accuracy of measurement model can be reaches above the98%. Secondly, in order to solve the shortage of real-time aspect for our measurementmodel, this paper proposes an optimized GPU strategy for the part of the algorithm.Through analysis of our study and experiments, we conclude the optimized object andmethod in the measurement algorithm model. The process of SIFT and SURF algorithmhas large image data processing demands. And the data processing is independent,———————————————————————————————————————————————— regularity, and structured, so we can speed up the part with the GPU. And through theoptimized GPU strategy, the SIFT can improve32times of the speed in CPU, and theGPU of the SURF can improve23times of the CPU’s speed. The measuring algorithm foran object based on multi-view images and its parallel optimization on GPUs can bequickly, accurately and effective achieve the target object height measurements.