Granted Patent B2

US 12,561,950 · App. 17/936,519 · Granted Feb 24, 2026

Inventors: Michele Merler (New York City, NY); Dhiraj Joshi (Edison, NJ); Apurv Gupta (Vadodara, IN); Sebastien Gilbert (Granby, CA); Shyama Prosad Chowdhury (West Bengal, IN); Chidansh Amitkumar Bhatt (Hightstown, NJ); Nirmit V. Desai (Yorktown Heights, NY)

Assignee: International Business Machines Corporation

G06V10/764G06V10/23G06V10/761G06V10/85G06V30/19173G06V2201/07

Abstract

Claims (73)

1 . A computer-implemented method of automated analog gauge reading, the method comprising:

receiving, by a computer system, input of an image;

detecting, by the computer system, at least one analog gauge in the image;

correcting, by the computer system, orientation of the at least one analog gauge in the image;

detecting, by the computer system, scene text and tick labels on the at least one analog gauge;

determining, by the computer system, a position of a pointer on the at least one analog gauge relative to the scene text; and

outputting, by the computer system, a gauge reading value based on an arithmetic progression of tick labels and angle of the pointer with respect to minimum and maximum values on the at least one analog gauge.

2 . The method of claim 1 , further comprising correcting, by the computer system, perspective distortion of the at least one analog gauge in the image to rectify the at least one analog gauge to a canonical viewpoint.

3 . The method of claim 2 , wherein correcting perspective distortion comprises:

converting, by the computer system, the input image to grayscale;

blurring, by the computer system, the image to remove high frequency noise;

identifying, by the computer system, candidate ellipse points in the image;

performing, by the computer system, ellipse fitting with the ellipse points according to random sample consensus (RANSAC); and

affinely transforming, by the computer system, the ellipse to a circle.

4 . The method of claim 1 , wherein determining the position of the pointer on the at least one analog gauge comprises:

performing, by the computer system, inner dial detection via dynamic thresholding on an intensity distribution on the image;

identifying, by the computer system, a highest probability inner component line in the image;

performing, by the computer system, co-linearity fitness analysis on regions geometrically localized in same neighborhoods in the image;

analyzing, by the computer system, a probable pointer in a neighborhood with fitness; and

performing, by the computer system, multilevel filtration and grouping on supporting lines in the image.

5 . The method of claim 1 , wherein detecting the at least one analog gauge in the image is performed with an object detection model trained to detect analog gauges.

6 . The method of claim 1 , wherein detecting scene text on the at least one analog gauge in the image is performed with an object detection model trained to detect number regions.

7 . The method of claim 1 , wherein correcting orientation of the at least one analog gauge in the image is performed with a rotation classification model trained on images with corresponding orientation information.

8 . A system for automated analog gauge reading, the system comprising:

a storage device that stores program instructions; and

one or more processors operably connected to the storage device and configured to execute the program instructions to cause the system to:

receive input of an image;

identify at least one analog gauge in the image;

correct orientation of the at least one analog gauge in the image;

detect scene text and tick labels on the at least one analog gauge;

determine a position of a pointer on the at least one analog gauge relative to the scene text; and

output a gauge reading value based on an arithmetic progression of tick labels and angle of the pointer with respect to minimum and maximum values on the at least one analog gauge.

9 . The system of claim 8 , further comprising correcting perspective distortion of the at least one analog gauge in the image to rectify the at least one analog gauge to a canonical viewpoint.

10 . The system of claim 9 , wherein correcting perspective distortion comprises:

converting the input image to grayscale;

blurring the image to remove high frequency noise;

identifying candidate ellipse points in the image;

performing ellipse fitting with the ellipse points according to random sample consensus (RANSAC); and

affinely transforming the ellipse to a circle.

11 . The system of claim 8 , wherein determining the position of the pointer on the at least one analog gauge comprises:

performing inner dial detection via dynamic thresholding on an intensity distribution on the image;

identifying a highest probability inner component line in the image;

performing co-linearity fitness analysis on regions geometrically localized in same neighborhoods in the image;

analyzing a probable pointer in a neighborhood with fitness; and

performing multilevel filtration and grouping on supporting lines in the image.

12 . The system of claim 8 , wherein:

detecting the analog gauge in the image is performed with an object detection model trained to detect analog gauges; and

detecting scene text on the at least one analog gauge in the image is performed with an object detection model trained to detect number regions.

13 . The system of claim 8 , wherein correcting orientation of the at least one analog gauge in the image is performed with a rotation classification model trained on images with corresponding orientation information.

14 . A computer program product for automated analog gauge reading, the computer program product comprising:

a persistent storage medium having program instructions embodied thereon to perform the steps of:

receiving input of an image;

identifying at least one analog gauge in the image;

correcting orientation of the at least one analog gauge in the image;

detecting scene text and tick labels on the at least one analog gauge;

determining a position of a pointer on the at least one analog gauge relative to the scene text; and

outputting a gauge reading value based on an arithmetic progression of tick labels and angle of the pointer with respect to minimum and maximum values on the at least one analog gauge.

15 . The computer program product of claim 14 , further comprising correcting perspective distortion of the at least one analog gauge in the image to rectify the at least one analog gauge to a canonical viewpoint.

16 . The computer program product of claim 15 , wherein correcting perspective distortion comprises:

converting the input image to grayscale;

blurring the image to remove high frequency noise;

identifying candidate ellipse points in the image;

performing ellipse fitting with the ellipse points according to random sample consensus (RANSAC); and

affinely transforming the ellipse to a circle.

17 . The computer program product of claim 14 , wherein determining the position of the pointer on the at least one analog gauge comprises:

performing inner dial detection via dynamic thresholding on an intensity distribution on the image;

identifying a highest probability inner component line in the image;

performing co-linearity fitness analysis on regions geometrically localized in same neighborhoods in the image;

analyzing a probable pointer in a neighborhood with fitness; and

performing multilevel filtration and grouping on supporting lines in the image.

18 . The computer program product of claim 14 , wherein detecting the at least one analog gauge in the image is performed with an object detection model trained to detect analog gauges.

19 . The computer program product of claim 14 , wherein detecting scene text on the at least one analog gauge in the image is performed with an object detection model trained to detect number regions.

20 . The computer program product of claim 14 , wherein correcting orientation of the at least one analog gauge in the image is performed with a rotation classification model trained on images with corresponding orientation information.

Assignments (1)

Continuity (1)

References Cited (31)