Applicable recognition system for estimating object conditions

• US 5,845,048 A
• Filed: 02/06/1996
• Issued: 12/01/1998
• Est. Priority Date: 02/06/1995
• Status: Expired due to Term

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Masumoto, Daiki
• Primary Examiner(s): Hafiz, Tariq R.
• Application Number:
  US08/597,138
• Time in Patent Office: 1,029 Days
• Field of Search: 382/291, 382/133, 382/48, 382/49, 382/14, 382/15, 382/41, 395/22, 395/21, 395/600, 395/20, 395/25, 395/27
• US Class Current: 706/20
• CPC Class Codes: B25J 9/1697 Vision controlled systems
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 18/21 Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06N 3/045 Combinations of networks

Abstract

A system for recognizing an object state (the kind, position, attitude) gets observation object information and recognizes a naturally occurring object for which simple rules cannot be found. This system outputs a state prediction value for a targeted object from cognitive object observation data, an observation data prediction value for targeted partial features according to this output, a partial feature prediction position in targeted observation results according to a state evaluation value output, and recognizes the object by modifying an output of the state prediction value output unit. Many naturally occurring objects, for which simple rules cannot be found, can be modeled, and a recognition system, which can be used in more realistic circumstances, can be constructed. An object based vision and an observer based vision as a 3D object model representation method is provided, and learning patterns necessary for recognizing are obtained effectively by using a visible-invisible determination module in automatically obtaining the 3D object model.

Claims

1. A system for recognizing an object from cognitive observation data existing in external circumstances, comprising:

state prediction value output means for outputting a state prediction value of said object from observation data for a cognitive object;

partial feature observation result prediction means for outputting a predicted value of said observation data for a partial feature of said cognitive object for input of said state prediction value, said partial feature observation result prediction means comprising more than one hierachical neural network each corresponding to more than one partial feature of said cognitive object;

partial feature prediction position output means for outputting an observation position of more than one partial feature in observation results of said cognitive object for said input of said state prediction value; and

means for recognizing said object by decreasing a difference between a predicted position which said partial feature prediction position output means outputs and a position on observation results for each partial feature determined based on said output from said partial feature observation result prediction means.

2. The system defined in claim 1, further comprising:

visible-invisible determination means for determining whether each one of more than one partial feature in said cognitive object is visible or invisible for said observation data of said cognitive object.

3. The system defined in claim 2, wherein

said state prediction value output means use a geometric hashing method when predicting cognitive object states.

4. The system defined in claim 2, wherein

said state prediction value output means use a combined method of a hough conversion method and a geometric hashing method when predicting cognitive object states.

5. The system defined in claim 2, wherein

said partial feature observation result prediction means comprise more than one neural network, each of which corresponds to each of more than one partial feature of said cognitive object.

6. The system defined in claim 5, wherein

a hierarchical neural network learning process comprises a preparation learning phase in which users teach which part corresponds to partial feature observation data in sensor data as cognitive object observation data, and an automatic learning phase in which a system automatically learns said sensor data corresponding to said partial features.

7. The system defined in claim 2, wherein

said partial feature prediction position output means comprise a mathematical model describing relationships between more than one partial feature prediction position and said state prediction value.

8. The system defined in claim 2, wherein

said partial feature prediction position output means comprise hierarchical neural networks.

9. The system defined in claim 2, wherein

a square error between an actual partial feature position and output from said partial feature prediction position output means is used as an estimation function, an output from said state prediction value output means is modified, and said object is recognized.

10. The system defined in claim 1, further comprising:

outside circumstance observation sensor means for giving cognitive object observation data existing in said outside circumstances to said state prediction value output means.

11. The system defined in claim 10, wherein

said outside circumstance observation sensor means comprise a visible sensor.

12. The system defined in claim 10, wherein

said outside circumstance observation sensor means comprise a plurality of kinds of sensors.

13. The system defined in claim 10, further comprising:

means for retrieving partial data containing more than one partial features of said cognitive object from output data from said outside circumstance observation sensor and giving to said state prediction value output means.

14. The system defined in claim 10, further comprising:

means for prior processing output data from said outside circumstance observation sensor, and giving all data or partial data which are prior processed to said state prediction value output means.

15. The system defined in claim 10, further comprising:

means for extracting more than one edge of said cognitive object from external circumstance observation sensor output data, and giving data for extracted edges to said state prediction value output means.

16. The system defined in claim 1, wherein

said state prediction value output means use a generalized Hough conversion when predicting states for said cognitive object.

17. The system defined in claim 1, wherein

said state prediction value output means use a geometric hashing method when predicting cognitive object states.

18. The system defined in claim 1, wherein

said state prediction value output means use a combined method of a geometric hashing method and a Hough conversion method when predicting cognitive object states.

19. The system defined in claim 1, wherein

a learning process for said hierarchical networks comprises a preparation learning phase in which users teach which part a partial feature corresponds to in sensor data as observation data of said cognitive object, and an automatic learning phase in which a system automatically learns sensor data corresponding to said partial feature.

20. The system defined in claim 1, wherein

said partial feature prediction position output means comprise a mathematical model describing relationships between a prediction position of more than one partial feature image and said state prediction value.

21. The system defined in claim 1, wherein

said partial feature prediction position output means comprise a hierarchical neural network.

22. The system defined in claim 1, wherein

a square error between an actual partial feature position and output from said partial feature prediction position output means is used as prediction function, an output from said state prediction value output means is modified in order to minimize said prediction function, and said object is recognized.

23. The system defined in claim 1, wherein

said state prediction value output means use a generalized Hough conversion method for predicting cognitive object states.