System and method of image-based space detection

• US 8,059,864 B2
• Filed: 04/28/2008
• Issued: 11/15/2011
• Est. Priority Date: 09/28/2007
• Status: Active Grant

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: Wu, Ruei-Cheng, Kuan, Cheng-Peng, Chang, Yao-Jen, Huang, Ching-Chun
• Primary Examiner(s): TUCKER, WESLEY J
• Application Number:
  US12/111,190
• Publication Number:
  US 20090085772A1
• Time in Patent Office: 1,296 Days
• Field of Search: 382103-105, 340/932.2, 340/933, 340/937, 701/1
• US Class Current: 382/103
• CPC Class Codes: G06F 18/2415 based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
• CPC Class Codes: G06F 18/29 Graphical models, e.g. Bayesian networks
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/84 using probabilistic graphical models from image or video features, e.g. Markov models or Bayesian networks
• CPC Class Codes: G06V 20/54 of traffic, e.g. cars on the road, trains or boats
• CPC Class Codes: G08G 1/14 indicating individual free spaces in parking areas

Abstract

Disclosed is a system and method of image-based space detection. The system includes an image selection module, a 3-layer detection mechanism and an optimization module. At least one image processing area that may affect space-status judgment is selected from plural image processing areas. The 3-layer detection mechanism having an observation layer, a labeling layer, and a semantic layer observes the information about the selected image processing area, associates with a local classification model, and adjacent local constraint model and a global semantics model to completely describe the probability distribution of the links among the three layers, and provide global label constraint information. The optimization module analyzes the probability distribution and global label constraint information, and generates an image-based optimized space detection result.

Claims

1. An image-based space detection system, comprising:

an image capturing and processing device for capturing and processing images of a plurality of image processing areas and selecting from said captured image processing areas at least an image processing area affecting space status determination;

a 3-layer detection mechanism having an observation layer, a labeling layer and a semantics layer, said 3-layer detection mechanism being coupled to said image capturing and processing device, observing information of said selected image processing areas, combining a local classification model, an adjacent local constraint model and a global semantics model to fully describe probability distribution of link relation between layers of said observation layer, said labeling layer and said semantic layer, and providing global labeling constraint information; and

a computing device connected to said 3-layer detection mechanism, for analyzing said probability distribution and said global labeling constraint information, and generating image-based optimal space detection result.

2. The system as claimed in claim 1, wherein said observation layer is an image observation layer, said labeling layer is a hidden labeling layer, and said semantics layer is an object semantics layer.

3. The system as claimed in claim 2, wherein each node in said image observation layer represents a local feature of an area.

4. The system as claimed in claim 2, wherein said each node in said hidden labeling layer represents classification labeling status of an area or a pixel.

5. The system as claimed in claim 2, wherein said each node in said object semantics layer comprises all the assumed statuses that said system is expected to derive.

6. The system as claimed in claim 2, wherein said link relation between said image observation layer and said hidden labeling layer is established through said local classification model.

7. The system as claimed in claim 2, wherein said link relation between nodes in said hidden labeling layer is established through said adjacent local constraint model.

8. The system as claimed in claim 2, wherein said link relation between said object semantics layer and said hidden labeling layer is established through said global semantics model.

9. The system as claimed in claim 2, wherein said local classification model classifies information of said selected at least an image processing area into different types of labeling.

10. The system as claimed in claim 1, wherein said local classification model, said adjacent local constraint model and said global semantics model are all pre-trained models.

11. The system as claimed in claim 1, wherein said local classification model, said adjacent local constraint model and said global semantics model are models that are based on pre-trained model and updated online according to detection result.

12. The system as claimed in claim 1, wherein said optimal space detection result is the space status, vacancy location, the number of vacancies, or any combination of the above.

13. The system as claimed in claim 1, wherein said global semantics model provides labeling constraint information of global semantics prior knowledge, and said labeling constraint information is hidden 3-D projection information, predictable occlusion information, or any combination of the above.

14. The system as claimed in claim 12, wherein all said parking spaces may be with a labeling, without any labeling, or some having labeling and some having not labeling.

15. The system as claimed in claim 1, wherein said system is a space detection system for parking lot.

16. The system as claimed in claim 15, wherein each classification labeling node of said hidden labeling layer represents one of a “

car”

classification labeling, a “

ground”

classification labeling, a “

special labeling”

classification labeling”

, or an “

other”

classification labeling.

17. The system as claimed in claim 10, wherein said adjacent local constraint model learns from a training set constraint relation between adjacent labeling nodes in classification adjacent nodes.

18. An image-based space detection method, comprising the steps of:

obtaining a plurality of image processing areas, and selecting from said obtained image processing areas at least an image processing area affecting space status determination by using an image capturing and processing device to capture and processing images of the plurality of image processing area;

establishing a 3-layer detection mechanism for observing information of said selected image processing area, combining a local classification model, an adjacent local constraint model and a global semantics model to fully describe probability distribution of link relation between layers of said 3-layer detection mechanism, and providing global labeling constraint information; and

analyzing said probability distribution and said global labeling constraint information, and generating image-based optimal space detection result by using a computing device.

19. The method as claimed in claim 18, wherein said optimal space detection result is obtained through an optimization inference method.

20. The method as claimed in claim 18, wherein said 3-layer detection mechanism is established as a 3-layer detection mechanism having an image observation layer, a hidden labeling layer and an object semantics layer.

21. The method as claimed in claim 18, said method further includes training a graphic model most suitable for said 3-layer detection mechanism, and said graphic model filly describes related statistics feature in said 3-layer detection mechanism and generates probability distribution of link relation between said layers.

22. The method as claimed in claim 19, wherein said optimization inference method accelerates inference process through the statistics relation between said object semantics layer and said hidden labeling layer.

23. The method as claimed in claim 19, wherein said optimization inference method finds the most suitable classification labeling and the optimal semantics inference according to observation information, and describes as an optimization mapping problem for solving.

24. The method as claimed in claim 19, wherein said optimization inference method is a local optimization method, and said local optimization method further includes:

finding an initial guess of image labeling according to current observation;

inferring space status according to said image labeling result;

re-calibrating said image labeling and inferring new space status according to current observed image information and said inferred space status; and

comparing new inferred space status and previously inferred space status, if the same, ending said inference process;

otherwise, returning to the step of re-calibrating said image labeling and inferring new space status.