Imaging, Robotics, and Intelligent Systems IRIS 3D Reconstruction of Tire-soil Interaction Wei Hao Imaging, Robotics, & Intelligent Systems Laboratory The University of Tennessee Mar 23, 2004 July 12, 2016 Slide 2 Possible Sponsor Needs Imaging, Robotics, and Intelligent Systems IRIS - Modeling the real world tire-soil interaction • Applying state of art computer vision techniques to generate accurate 3D model of the real world tire-soil interaction. - Our models can be used for real-time measurement of the tire deformation • With the 3D modeling of the tire deformation, it is easy to get real data for tire design. - Our method is a new perspective to tire-soil interaction simulation problem • efforts of establish analytical models of the tire-soil interaction has been made by several research groups; • Our method will provide real data set of the interaction which can be used as validation model. -The method proposed here is applicable for all kinds of soil • Analytical models for soft soil and general terrains are very difficult to be established; • Lab data set based on mechanical sensors are also not easily obtained; • Fast digitizing of the tire-soil interaction provides accurate models of the whole process. July 12, 2016 Slide 3 Imaging, Robotics, and Intelligent Systems Related Research Work IRIS - Modeling and simulation of real time tire-soil interaction is demanded from military and defense industries, mainly for mobility analysis. - Researchers has built numerical models for tire and terrain based on techniques such as finite element analysis. Integration of tire and terrain dynamic models constitutes the tiresoil interaction model. kinematic tire models are built with wheel slip ratio, slip angle and normal force in consideration. - The kinematic models are numerical simulation rather than real world. - Computer vision and photogrammetry techniques are matured enough to provide accurate and real time digitizing of real world tire-terrain interaction, which cannot be achieved by mechanical analysis and numerical modeling. [Lee etal 05] Lee, J.H., Liu, Q., and Zhang, T., “Predictive Semi-Analytical Model for Tire-Snow Interaction”, SAE2005-01-0932, 2005 [Pan 05] W. Pan, “Modeling Tire Transient Response in Real-Time Simulation”, NSF I/UCRC Center for Virtual Proving Ground 17th Semiannual Conference, Iowa City, Iowa, April 5-6, 2005 [Wal etal 04]Walsh, SDC and Tordesillas, “A thermomechanical approach to the development of micropolar constitutive models for granular media” Acta Mechanica, 167 (3-4), 2004 July 12, 2016 Slide 4 3D Reconstruction of Tire-Soil Interaction Imaging, Robotics, and Intelligent Systems IRIS Motivations: Digitize the tire-soil interaction and build a 4D movie of it (3D reconstruction time series) using state-of-arts techniques of computer vision. Fig 1. Typical scene of tire-soil interaction to be modeled. (Courtesy of US ARMY) Objectives - Provide new method of tire-soil interaction study other than mechanism modeling. - Build automated data acquisition system which can digitize the tire-soil interaction process in real time; - Develop robust algorithms which can generate time series of 3D reconstruction; - Build models of soil-tire interaction for on-road and off-road and test them using real data. July 12, 2016 Slide 5 Diagram of Data Acquisition System Imaging, Robotics, and Intelligent Systems IRIS Data Collection System GUI Date Collection Program Camera synchronization control Console/Computer Frame Grabber Software Camera 1 Camera 2 Fig 2. Data Acquisition System Diagram. … Camera n Hardware July 12, 2016 Slide 6 Data Acquisition System Implementation Stereo Vision System Sensor Placement Video of Data Acquisition (1) Video of Data Acquisition (2) Imaging, Robotics, and Intelligent Systems IRIS July 12, 2016 Slide 7 Imaging, Robotics, and Intelligent Systems Sample Data Collections(1) • We collected stereo image sequences under difference circumstances. Fig 3. Stereo image sequence of tire running on soft soil. IRIS July 12, 2016 Slide 8 Imaging, Robotics, and Intelligent Systems Sample Data Collections(2) Fig 4. Stereo image sequence of tire running on grassy soil. IRIS July 12, 2016 Slide 9 Imaging, Robotics, and Intelligent Systems Sample Data Collections(3) Fig 5. Stereo image sequence of tire running on soft soil. IRIS July 12, 2016 Slide 10 Imaging, Robotics, and Intelligent Systems Sample Data Collections(4) Fig 6. Stereo image sequence of tire running on gravel road. IRIS July 12, 2016 Slide 11 Imaging, Robotics, and Intelligent Systems 3D Reconstruction Software Camera Calibration Image Rectification IRIS - Matured 3D reconstruction techniques available for static scene reconstruction; - We are now developing advanced temporal stereo techniques to generate 3D model sequence. Dense Matching Disparity Estimation Triangulation & Visualization Fig 7. 3D reconstruction based on multiple view geometry July 12, 2016 Slide 12 Imaging, Robotics, and Intelligent Systems Stereo Camera Calibration IRIS -The relative translation and rotation of the two cameras are constant during the data acquisition processing. -Camera calibration method: [Zhang 1998] Fig 8. Frame 10, 12,13,14 of the different pose of the planar calibration rig of the left camera. [Zhang 1998] Z. Zhang. A Flexible New Technique for Camera Calibration. Technical Report MSRTR-98-71, Microsoft Research, Dec 1998. July 12, 2016 Slide 13 Imaging, Robotics, and Intelligent Systems IRIS Stereo Camera Calibration Calibration Results: The intrinsic parameters for the two stereo cameras and the relative displacement of the two cameras. Stereo calibration parameters: Intrinsic parameters of left camera: Focal Length: fc_left = [ 1135.33835 1134.75205 ] ± [ 14.29820 13.98266 ] Principal point: cc_left = [ 518.01575 371.44273 ] ± [ 23.44983 16.16967 ] Skew: alpha_c_left = [ 0.00000 ] ± [ 0.00000 ] Distortion: kc_left = [ -0.01518 -0.27005 -0.00295 -0.01034 0.00000 ] ± [ 0.04661 0.00000 ] 0.33772 0.00391 0.00641 0.29916 0.00383 0.00316 Intrinsic parameters of right camera: Focal Length: fc_right = [ 1138.66134 1141.58652 ] ± [ 13.88601 13.73518 ] Principal point: cc_right = [ 488.89808 326.89759 ] ± [ 15.09664 16.76655 ] Skew: alpha_c_right = [ 0.00000 ] ± [ 0.00000 ] Distortion: kc_right = [ -0.14638 0.53836 -0.01497 -0.00728 0.00000 ] ± [ 0.04115 0.00000 ] Extrinsic parameters (position of right camera wrt left camera): Rotation vector: Translation vector: om = [ -0.05570 0.01084 -0.00898 ] ± [ 0.01026 0.02087 0.00129 ] T = [ -102.38391 4.02867 4.61723 ] ± [ 0.46797 0.32744 3.34022 ] July 12, 2016 Slide 14 Relative Pose of the Calibration Rig with Respect to the Cameras Fig 9. Position of All the 6*6 feature regions w.r.t the stereo system computed. Imaging, Robotics, and Intelligent Systems IRIS July 12, 2016 Slide 15 Imaging, Robotics, and Intelligent Systems Image Rectification IRIS - Rectification can simplify the dense matching process. After rectification, the correspondence will be of the same x-coordinates. - Use the method of [Fusiello etal 2000]. It require good synchronization between the calibration image pairs. Fig 10. Rectified Frame 10, 12,13,14 of the left camera. [Fusiello etal 2000] Andrea Fusiello, Emanuele Trucco, Alessandro Verri. “A compact algorithm for rectification of stereo pairs”, Machine Vision and Applications(2000) 12:16-22 July 12, 2016 Slide 16 Imaging, Robotics, and Intelligent Systems IRIS Dense Matching Stereo matching methods: Local Matching Methods: Global Methods: -Block Matching; -Dynamic Programming; -Gradient-based Optimization -Intrinsic Curves -Feature Matching -Graph Cuts -Nonlinear Diffusion -Belief Propagation -Correspondenceless Methods July 12, 2016 Slide 17 Imaging, Robotics, and Intelligent Systems Sample Dense Matching Results Sample inputs captured by the same data acquisition set: Fig 11. Sample inputs of stereo images IRIS July 12, 2016 Slide 18 Imaging, Robotics, and Intelligent Systems Dense Matching Outputs (a) Fig 12. Sample output of (a) SAD method (b) (b) SSD+DP Method IRIS July 12, 2016 Slide 19 Imaging, Robotics, and Intelligent Systems Temporal Stereo Techniques IRIS Stereo vision and structure from motion are two well formed techniques for 3D reconstruction. In current computer vision research, combining both spatial and temporal image information is becoming increasingly popular as a way of improving the robustness and computational efficiency of recovering the 3-D structure of a scene. Related work: [Dav03] J. Davis, R. Ramamoorthi, and S. Rusinkiewicz. Spacetime stereo: A unifying framework for depth from triangulation. In CVPR, 2003. [Zha 03] Li Zhang, Brian Curless, and Steven M. Seitz. Spacetime Stereo: Shape Recovery for Dynamic Scenes. In Proceedings of CVPR), Madison, WI, June, 2003, pp. 367-374. [Leu 04] Leung, C. and Appleton, B. and Lovell, Brian C. and Sun, C. An Energy Minimisation Approach to Stereo-Temporal Dense Reconstruction. In ICPR 2004 July 12, 2016 Slide 20 Imaging, Robotics, and Intelligent Systems Future Work - Add more sensors into the data acquisition system; - collect more data set for different soil; - Finish the software and generate the 3D models and the 4D movie. IRIS