Image-based hand detection apparatus and method

• US 8,638,987 B2
• Filed: 09/22/2011
• Issued: 01/28/2014
• Est. Priority Date: 10/06/2010
• Status: Active Grant

• Current Assignee: Electronics and Telecommunications Research Institute
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Choi, Seung-Min, Jeong, Jae-Chan, Cho, Jae-Il, Hwang, Dae-Hwan
• Primary Examiner(s): Patel, Jayesh A
• Assistant Examiner(s): Kholdebarin, Iman K
• Application Number:
  US13/241,119
• Publication Number:
  US 20120087543A1
• Time in Patent Office: 859 Days
• Field of Search: 382/103, 382/190, 382/209, 382/216
• US Class Current: 382/103
• CPC Class Codes: G06V 40/11 Hand-related biometrics; Hand pose recognition

Abstract

An image-based hand detection apparatus includes a hand image detection unit for detecting a hand image corresponding to a shape of a hand clenched to form a fist from an image input. A feature point extraction unit extracts feature points from an area, having lower brightness than a reference value, in the detected hand image. An image rotation unit compares the feature points of the detected hand image with feature points of hand images stored in a hand image storage unit, and rotates the detected hand image or the stored hand images. A matching unit compares the detected hand image with the stored hand images and generates a result of the comparison. If at least one of the stored hand images is matched with the detected hand image, a hand shape recognition unit selects the at least one of the stored hand images as a matching hand image.

Claims

1. An image-based hand detection apparatus, comprising:

a hand image detection unit configured to detect a hand image corresponding to a shape of a hand clenched to form a fist from an image input from a camera;

a feature point extraction unit configured to extract feature points from an area, having lower brightness than a reference value, in the hand image detected by the hand image detection unit;

an image rotation unit configured to compare the feature points of the detected hand image with feature points of hand images stored in a hand image storage unit, and then rotate the detected hand image or the stored hand images;

a matching unit configured to compare the detected hand image with the stored hand images and then generate a result of the comparison; and

a hand shape recognition unit configured to, if at least one of the stored hand images is matched with the detected hand image as the result of the comparison by the matching unit, select the at least one of the stored hand images as a matching hand image,

wherein the feature point extraction unit is further configured to extract the number of edges, locations of the edges, and angles of the edges as the feature points of the detected hand image, the edges corresponding to areas with lower brightness than the reference value in the detected hand image.

2. The image-based hand detection apparatus of claim 1, wherein the hand image detection unit is further configured to set a window of an M×

N size on a pixel unit for each frame of the input image, and detect the hand image of the M×

N size.

3. The image-based hand detection apparatus of claim 2, wherein the hand image detection unit is further configured to determine whether a hand is included in the M×

N size window of the input image, and if the hand is not included in the M×

N size window, reduce a size of the input image step by step for detecting the hand image in the M×

N size window.

4. The image-based hand detection apparatus of claim 1, wherein the hand shape recognition unit is further configured to recognize a front portion of the shape of the hand clenched to form a fist.

5. The image-based hand detection apparatus of claim 1, further comprising a hand shape classifier configured to, if all the stored hand images are not matched with the detected hand image as the result of the comparison by the matching unit, register the detected hand image.

6. The image-based hand detection apparatus of claim 5, wherein the hand shape classifier is further configured to compare a plurality of hand images with images including no hands, calculate feature values of a hand shape included in each of the hand images, and classify each hand shape based on the calculated feature values.

7. The image-based hand detection apparatus of claim 6, wherein the hand shape classifier is configured such that if an identical hand shape is present in hand shapes included in the respective hand images, weights are applied to feature values of the identical hand shape, and then feature values are recalculated.

8. The image-based hand detection apparatus of claim 1, further comprising a movement recognition unit configured to recognize hand movements from one or more hand shapes recognized by the hand shape recognition unit and recognize commands corresponding to the recognized hand movements.

9. An image-based hand detection method, comprising:

detecting a hand image corresponding to a shape of a hand clenched to form a fist from an image input from a camera, and generating a detected hand image;

extracting feature points from an area, having lower brightness than a reference value, in the detected hand image;

comparing the feature points of the detected hand image with feature points of hand images stored in a hand image storage unit to rotate the detected hand image or the stored hand images, and comparing the detected hand image with the stored hand images to generate a result of the comparison; and

if at least one of the stored hand images is matched with the detected hand image as the result of the comparison, selecting the at least one of the stored hand images as a matching hand image,

wherein the extracting the feature points includes extracting the number of edges, locations of the edges, and angles of the edges as the feature points of the detected hand image, the edges corresponding to areas with lower brightness than the reference value in the detected hand image.

10. The image-based hand detection method of claim 9, wherein the detecting the hand image comprises setting a window of an M×

N size on a pixel unit for each frame of the input image, wherein the hand image of the M×

N size is detected.

11. The image-based hand detection method of claim 10, wherein the detecting the hand image further comprises:

determining whether a hand is included in the M×

N size window of the input image; and

if the hand is not included in the M×

N size window, reducing a size of the input image step by step for detecting the hand image in the M×

N size window.

12. The image-based hand detection method of claim 9, wherein the selecting the at least one of the stored hand images as the matching hand image is configured to recognize a front portion of the shape of the hand clenched to form a fist.

13. The image-based hand detection method of claim 9, further comprising, if all the stored hand images are not matched with the detected hand image as the result of the comparison, inputting the detected hand image to a hand shape classifier and then registering the detected hand image.

14. The image-based hand detection method of claim 13, wherein the registering the hand image comprises:

comparing a plurality of hand images with images including no hands, and calculating feature values of a hand shape included in each of the hand images;

setting the calculated feature values as feature values of the hand shape included in each of the hand images; and

classifying each hand shape based on the feature values of the hand shapes.

15. The image-based hand detection method of claim 14, further comprising:

if an identical hand shape is present in hand shapes included in the respective hand images, applying weights to feature values of the identical hand shape, and then recalculating feature values; and

setting the recalculated feature values as final feature values of the identical hand shape.

16. The image-based hand detection method of claim 9, further comprising:

recognizing hand movements from hand shapes included in one or more hand images selected as the matching hand image; and

recognizing commands corresponding to the hand movements.