- 1. Neural object classification by pattern recognition of one dimensional dataarrays which represent object information transformed by nonlinear functions. 1 Neural object classification by pattern recognition of one dimensional data arrays which represent object information transformed by nonlinear functions. presented by Kayhan Ince Thesis Supervisor: Univ.Prof. Dipl.-Ing. Dr.techn. FAVRE-BULLE, Bernard Thesis Co-Advisor: Dipl.-Ing. Fauaz Labadi ACIN – Automation and Control Institute
- 2. Outline 2 Outline lDefinition and aim lState of the art in neural pattern recognition lShape analysis (classification) and grasping process lSimulation Results lConclusion
- 3. Definition and aim 3 Definition and aim lPattern lGroups of measurements or observations, defining points in an appropriate multidimensional space. lPattern recognition lAims to classify data (patterns) based on either a priori knowledge or on statistical information extracted from the patterns. lMy aim is to classify 2-dimensional objects with the help of manipulator and by the use of pattern recognition.
- 4. Grasping Process 4 Grasping Process 7-DOF manipulator performs with whole arm grasping of a planar object Hyper-redundant manipulator - Serial-Chain-Mechanism - Planar, rotational Joints - n Degrees of Freedom (DOF) - Array of angles joints Base Link
- 5. Slide72 5 Classification process Classification shape = [pi/4 pi/4 pi/4 pi/4 pi/4 pi/4 pi/4 pi/4 pi/4 pi/4 pi/4 pi/4 pi/4 pi/4] FEATURE VECTOR DRAWTESTSET TURNINGFUNCTION PERIODIC CLASSIFICATION RESULTS CLASSIFY
- 6. Slide78 6 Fourier Descriptors Turning Functions A special property of Fourier descriptors is that a shape’s symmetry shows up in the feature vector. Arkin published an efficient method for comparing polygonal shapes. The notion of the turning function which represents the shape of an object. Classification PNN
- 7. Neural Object Classification 7 Neural Object Classification Artificial neuron Radial basis function Input Neuron 1 x ( ) u F() 2 x u y å + = b w x u i i 1 w 2 w b Output(classification)
- 8. Slide80 8 •A probabilistic neural network structure is able to classify the objects for tentacle case problem •Turning functions carry the distance information of the objects Conclusion
- 9. Thank you for your attention. 9 Thank you for your attention.

Thank you for your comment.