Method and apparatus for detecting changes in dynamic images

• US 5,548,659 A
• Filed: 09/23/1994
• Issued: 08/20/1996
• Est. Priority Date: 12/27/1991
• Status: Expired due to Term

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Okamoto, Yasukazu
• Primary Examiner(s): Mancuso, Joseph
• Assistant Examiner(s): DEL ROSSO, GERARD DOMNICK
• Application Number:
  US08/312,049
• Time in Patent Office: 697 Days
• Field of Search: 382/130, 382/275, 382/294, 382/276, 382/107, 382/254, 348/154, 348/155, 348/700
• US Class Current: 382/107
• CPC Class Codes: G06T 7/254 involving subtraction of images

Abstract

A method and apparatus for detecting changes in dynamic images. The method and apparatus are capable of detecting movement at a high precision of a moving object against a static background, regardless of the variation of the image taking environmental condition, such as a lighting condition. In the apparatus, input images of a moving object are taken against a static background sequentially, and difference images are obtained from successive ones of the input images. Then, an appropriate noise model for a lightness variation due to an image taking environmental condition under which the input images are taken is estimated according to the input images and the difference images. Changed regions in an entire imaging view field are detected according to the input images, the difference images, and the appropriate noise model.

Claims

1. An apparatus for detecting a moving object in dynamic images, comprising:

image input means for sequentially taking input images of the moving object against a static background;

difference calculation means for obtaining difference images from successive input images taken by said image input means as picture elements;

noise model estimation means for estimating an appropriate noise model representing a lightness variation due to an image taking environmental condition under which the input images are taken by the image input means, the estimation being made according to the input images taken by the image input means and the difference images obtained by the difference calculation means by estimating values of noise model parameters in a regression model, and specifying a ratio of a difference in lightness at each picture element of each difference image with respect to an average difference in lightness over an entire imaging view field of each difference image as a function of lightness at each picture element in each input image expressed in terms of the noise model parameters;

change determining means for determining changed regions in the entire imaging view field by compensating for the lightness variation by subtracting from the difference images obtained by the difference calculation means an estimated lightness variation based on the noise model estimated by the noise model estimation means; and

detecting means for detecting the moving object in the dynamic images according to the changed regions determined by the change determining means.

2. The apparatus of claim 1, wherein the appropriate noise model estimated by the noise model estimation means indicates the light variation due to the image taking environmental condition resulting from a flickering of a lighting system under which the input images are taken by the image input means.

3. The apparatus of claim 1, wherein the change determining means determines the changed regions as those regions which are associated with lightness changes not accounted for by the appropriate noise model estimated by the noise model estimation means as the lightness variation due to the image taking environmental condition.

4. The apparatus of claim 1, wherein the values of the noise model parameters are statistically estimated from a plurality of image data extracted from a plurality of the input images taken by the image input means successively and a plurality of the difference images obtained from said plurality of the input images by the difference calculation means, each image data for each picture element containing the lightness at each picture element of each input image, the difference at each picture element of each difference image, and the average difference of each difference image over the entire imaging view field in each difference image.

5. The apparatus of claim 4, wherein the values of the noise model parameters are statistically estimated by using a least square method.

6. The apparatus of claim 1, wherein the noise model estimation means estimates the values of the noise model parameters by:

calculating a normalized difference for each picture element by subtracting a difference estimation estimated according to the regression model by using the lightness at each picture element of each input image and the average difference over the entire imaging view field in each difference image, from an actual difference at each picture element of each difference image;

calculating a smoothed normalized difference for each picture element by smoothing the normalized difference for each picture element with respect to the normalized differences for neighboring picture elements of said each picture element; and

obtaining a binarized image according to a comparison of a square sum average of the smoothed normalized difference for each picture element and a predetermined threshold, to distinguish picture elements associated with lightness changes due to a movement of the moving object from picture elements associated with the static background.

7. The apparatus of claim 6, wherein the predetermined threshold represents a confidence interval for a distribution of the square sum averages of the smoothed normalized differences for all of the picture elements in each difference image at a desired confidence level.

8. The apparatus of claim 6, wherein the noise model estimation means obtains the binarized image by:

comparing the square sum average of the smoothed normalized difference for each picture element with the predetermined threshold;

labelling each picture element which is surrounded by neighboring picture elements having the square sum averages greater than the predetermined threshold;

deleting any minute labelled region formed by the labelled picture elements which has an area not greater than a prescribed area; and

detecting each labelled region formed by the labelled picture elements which has an area greater than the prescribed area as the changed regions.

9. A method for detecting a moving object in dynamic images, comprising the steps of:

(a) sequentially taking input images of the moving object against a static background;

(b) obtaining difference images as picture elements from successive input images taken at the step (a);

(c) estimating an appropriate noise model representing a lightness variation due to an image taking environmental condition under which the input images are taken at the step (a), according to the input images taken at the step (a) and the difference images obtained at the step (b) by estimating values of noise model parameters in a regression model, and specifying a ratio of a difference in lightness at each picture element of each difference image with respect to an average difference in lightness over an entire imaging view field of each difference image as a function of lightness at each picture element in each input image expressed in terms of the noise model parameters;

(d) determining changed regions in the entire imaging view field by compensating for the lightness variation by subtracting from the difference images obtained at the step (b) an estimated lightness variation based on the noise model estimated at the step (c); and

(e) detecting the moving object in the dynamic images according to the changed regions determined at the step (d).

10. The method of claim 9, wherein the appropriate noise model estimated at step (c) indicates the light variation due to the image taking environmental condition causing a flickering of a lighting system under which the input images are taken at the step (a).

11. The method of claim 9, wherein the changed regions are determined at the step (d) as those regions which are associated with lightness changes not accounted for by the appropriate noise model estimated at the step (c) as the lightness variation due to the image taking environmental condition.

12. The method of claim 9, wherein the values of the noise model parameters are statistically estimated from a plurality of image data extracted from a plurality of the input images taken at the step (a) successively and a plurality of the difference images obtained from said plurality of the input images at the step (b), each image data for each picture element containing the lightness at each picture element of each input image, the difference at each picture element of each difference image, and the average difference of each difference image over the entire imaging view field in each difference image.

13. The method of claim 12, wherein the values of the noise model parameters are statistically estimated by using a least square method.

14. The method of claim 9, wherein the values of the noise model parameters are determined by the steps of:

(cl) calculating a normalized difference for each picture element by subtracting a difference estimation estimated according to the regression model by using the lightness at each picture element of each input image and the average difference over the entire imaging view field in each difference image, from an actual difference at each picture element of each difference image;

(c2) calculating a smoothed normalized difference for each picture element by smoothing the normalized difference for each picture element calculated at the step (c1) with respect to the normalized differences for neighboring picture elements of said each picture element calculated at the step (c1); and

(c3) obtaining a binarized image according to a comparison of a square sum average of the smoothed normalized difference for each picture element calculated at the step (c2) with a predetermined threshold, to distingish picture elements associated with lightness changes due to a movement of the moving object from picture elements associated with the static background.

15. The method of claim 14, wherein at the step (c3), the predetermined threshold represents a confidence interval for a distribution of the square sum averages of the smoothed normalized differences for all the picture elements in each difference image at a desired confidence level.

16. The method of claim 14, wherein the step (c3) further comprises the steps of:

(c31) comparing the square sum average of the smoothed normalized difference for each picture element calculated at the step (c2) with the predetermined threshold;

(c32) labelling each picture element which is surrounded by neighboring picture elements which are judged to have square sum averages greater than the predetermined threshold by the comparison at the step (C31);

(c33) deleting any minute labelled region formed by the picture elements labelled at the step (c32) which has an area not greater than a prescribed area; and

(c34) detecting each labelled region formed by the picture elements labelled at the step (c32) which has an area greater than the prescribed area as the changed regions.