Stereo Vision Based Distance Measurement And Its Application

In computer graphics, Stereo Vision has been a research hotspot formany years, and it is been widely used in many areas. Two main problemsneed to be solved: camera calibration and stereo matching.After years of development some achievements have been made incamera calibration and stereo matching. For camera calibration we can usethe tool box in Matlab, but it worked not very well for far distancecalibration. In consider that we will calculate distance above55meters wemust solve far distance camera calibration at first. For stereo matching,accuracy and efficiency are always a pair of contradictory requirements. Weshould guarantee that the stereo matching algorithm has these two propertiesif we want to apply the algorithm into practice, so we should balancebetween efficiency and accuracy.In this article we focuses on stereo vision based distance measurementand its application and two problems solved in this article. We can get goodcalibration result through marking corresponding point on left and rightimage and grounded for stereo matching. For stereo matching we proposeAdvanced-Census algorithm which is a good combination of SAD andCensus and it has advantages of both. But we found the algorithm could notget real-time performance, so we speed up Advanced-Census in algorithmapplication section. We consider mainly two aspect－complexity andconcurrency. For complexity, we found that there is no need to compute allthe pixel, so we add edge detection before distance computation. Wedecrease time from3minutes to2seconds through this method. On theother hand, we found that the algorithm can compute the distance ofdifferent pixels simultaneously, so we use multi-thread technique such asGPU. We implemented dual-thread on PC and decrease the time consuming to1second. Finally, we apply this algorithm on vehicle systems and give ademo.