Skip to Main Content
 

Global Search Box

 
 
 
 

ETD Abstract Container

Abstract Header

Visually guided tactile and force-torque sensing for object recognition and localization

Rafla, Nader Iskander

Abstract Details

1991, Doctor of Philosophy, Case Western Reserve University, Electrical Engineering.
A model-based object recognition and localization system has been developed to recognize and locate three dimensional solid objects using vision range images and touch data. This system performs three main tasks: extraction of surface characteristics, integration of vision and touch data, and object recognition and localization. Surface characteristics of the sensed object are extracted from vision range images and touch data independently. From vision range images, surface normals are calculated. Points which have similar properties of their normals are grouped into regions called surface patches. Touch data is acquired using a probe gripped by two imaging tactile and force-torque sensors (Lord Corporation gripper SE-680 and Lord Corporation sensors LTS-200) mounted in a five degree of freedom robot arm (Intelledex 605T). The operation of this tactile system is similar to that of a commercial robot, q.v. a coordinate measuring machine. From touch data, normals are determined using feedback from the tactile and force-torque sensors to orient the probe normal to the surface. Surfaces are located using the inverse kinematics of the robot probe. The extracted vision and touch features are combined into vision-touch surface patches on the basis of surface normals and position. These vision-tou ch patches are processed for classification and surface equation determination. The surface equations are calculated for each surface patch using a least-square minimization method that used only a few touch and vision surface points within the same patch. A series of experiments was done to test the different components of the system individually and as a system for object recognition and localization. The touch data was acquired from a variety of physical objects. Vision range images corresponding to these physical objects were computer generated. Normally distributed noise was added to the vision range images to simulate errors due to timing jitter, etc. The vision and touch data analysis successfully extracted normals for planar, cylindrical, and spherical surfaces. The resulting surface features were integrated into vision-touch surface patches. Some vision points on cylindrical and spherical surfaces were not integrated into the vision-touch surface patches due to the noise in the vision images but all vision points were correctly identified for planar surfaces
Francis Merat (Advisor)
156 p.

Recommended Citations

Citations

  • Rafla, N. I. (1991). Visually guided tactile and force-torque sensing for object recognition and localization [Doctoral dissertation, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1059578946

    APA Style (7th edition)

  • Rafla, Nader. Visually guided tactile and force-torque sensing for object recognition and localization. 1991. Case Western Reserve University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=case1059578946.

    MLA Style (8th edition)

  • Rafla, Nader. "Visually guided tactile and force-torque sensing for object recognition and localization." Doctoral dissertation, Case Western Reserve University, 1991. http://rave.ohiolink.edu/etdc/view?acc_num=case1059578946

    Chicago Manual of Style (17th edition)