The Hough Transform for Vertical Object Recognition in 3D Images Generated from Airborne Lidar DataSlide 2Hough transform- based approach for detecting vertical objects of cylindrical shape:Slide 4Slide 53D Binary Edge ImagesHT Cylinder Detection Algorithm:Slide 8Comparison of radii & axes locations of HT-detected cylinders with field-surveyed data:The Hough Transform for Vertical Object Recognition in 3D Images Generated from Airborne Lidar DataChristopher ParrishECE533 ProjectDecember 2006GPS Reference Station Airborne LidarAirport Obstruction SurveyingLidar Point CloudVoxelize3D Grayscale Intensity Image3D Sobeloperator3D Grayscale Edge ImageThreshold segmentation3D Binary Edge ImageHough Transform to identify vertical cylindersVertical objects of interestHough transform- based approach for detecting vertical objects of cylindrical shape:3D Grayscale Image2D Color Image Laser Point Cloud TTzyxzfyfxfGGGf222zyxGGGf f242000242,484000484,242000242:xG202404202,404808404,202404202:yG242484242,000000000,242484242:zGotherwise0 if 1),,(Tf(x,y,z)zyxgGradient of a 3D image, f(x,y,z): Magnitude of the gradient: 3D Sobel operator (three 3x3x3 filters expressed here as sets of three 2D matrices) Thresholded (binary) edge image Computing Binary Edge Image:3D Binary Edge ImagesHT Cylinder Detection Algorithm:Input = 3D binary edge imageQuantize 3D parameter space. Initialize all accumulator cells to zero. For each nonzero voxel in 3D binary edge image, step through all values of s and t. At each location:Solve for r Round r to its nearest accumulator cell valueIncrement counter for that (s,t,r) accumulator cell.Find entry in 3D accumulator array with highest # of votes. Assume cylinders are vertical (axes parallel to mapping frame Z axis) => # of parameters reduced from 5 to 3. Representation: (X-s)2+(Y-t)2 = r2Cylinders Detected Using Hough Transform:Comparison of radii & axes locations of HT-detected cylinders with field-surveyed
View Full Document