Research On Three-dimensional Measurement Of Material Stack Based On Stereo Vision

In the development of urban infrastructure construction and some large-scale projects,such as dam construction,mining,port storage and transportation,these engineering scenarios will appear fine sand pile,scree and other material heaps.The quantification of these material heaps is a very important link for the engineering scenarios,and the effective measurement of the material heap can improve the management of the construction project and improve the work efficiency.However,there is not very suitable measurement method for these bulky and irregular material heaps;in the past five years,with the comprehensive study of stereo vision technology,more and more intelligent automobile,exploration vehicles and other vehicles are equipped with stereo vision sensors to detect the surrounding environment.Based on the above reasons,the thesis proposes a method to obtain three-dimensional information of material heap to estimate the volume of the target from the perspective of vehicle stereo vision,mainly from the following three aspects:The feature point extraction problem of calibrated stereo vision image was studied,aiming at the problem of the number of feature points and the speed of the extraction algorithm,the thesis combines FAST algorithm with Canny algorithm with better effect of edge feature extraction.,merging the feature points by the two algorithms in the established scale space to enhance the feature information of edge points.Since none of the two algorithms have rotational invariance and scale invariance,the description method of feature points in SIFT algorithm is applied to the feature points,which are extracted by two algorithms.The algorithm was tested and compared in different scenarios,the algorithm proposed in this thesis can get more correct matching feature points pairs and has been improved accordingly.After a good extraction method for the feature information of the plane in the binocular images,it was necessary to obtain the depth information of the target from the left and right images.Compared with two different methods of depth image acquisition,local stereo matching BM algorithm and semi-global stereo matching SGBM algorithm in this thesis.Both algorithms have good effect on extracting depth information on the target object with obvious feature information,but when the target is material heap,SGBM has advantages in effect of generating dense disparity map.After obtaining the plane feature information and the target object depth information,gaining the coordinate points of the three dimensional space from the plane feature information and the depth information.The profile of the target object is fitted by these coordinate points,and the volume of the target object is estimated.Finally,an experimental platform is set up to simulate the relationship between the engineering vehicle and the material heap.A series of experiments are designed to verify the effect of this algorithm in this thesis on the three-dimensional information extraction of the material heap.Compared the influence of materials heap in different materials on the algorithm and the influence of material heap on the algorithm when the volume changes occur.The measurement accuracy of the algorithm is evaluated and the errors are analyzed in detail that will be produced in the whole measurement process.