Illumination invariant and robust apparatus and method for detecting and recognizing various traffic signs

• US 9,269,001 B2
• Filed: 06/10/2011
• Issued: 02/23/2016
• Est. Priority Date: 06/10/2010
• Status: Active Grant

• Current Assignee: TATA Consultancy Services Limited (Tata Sons Pvt Ltd.)
• Original Assignee: TATA Consultancy Services Limited (Tata Sons Pvt Ltd.)
• Inventors: Pulugoru, Pratap, Sinha, Aniruddha, Agsebagil, Sachin, R, Manoj C
• Primary Examiner(s): Harold, Jefferey
• Assistant Examiner(s): Khalid, Omer
• Application Number:
  US13/703,052
• Publication Number:
  US 20130201334A1
• Time in Patent Office: 1,719 Days
• Field of Search: 382/164, 382/103, 382/104, 382/167, 348/148, 348/149
• US Class Current: 1/1
• CPC Class Codes: G06F 2218/12 Classification; Matching
• CPC Class Codes: G06T 5/00 Image enhancement or restoration
• CPC Class Codes: G06V 20/582 of traffic signs
• CPC Class Codes: H04N 23/70 Circuitry for compensating brightness variation in the scene

Abstract

The present application provides a robust, illumination invariant apparatus and method for detecting and recognizing various traffic signs. A robust method for detecting and recognizing the traffic signs using images captured by a digital color and night vision camera, the said method characterized in being illumination invariant comprising the processor implemented steps of: transforming RGB image into HSV color model and subsequently extracting desired color components by using color quantization; filtering the noise components in the HSV color model based on object symmetrical shape property; detecting edges of the objects and subsequently detecting the distinct objects in the noise components filtered image; classifying the shapes of the traffic signs based on shape of the determined distinct objects; and recognizing the classified shapes of the traffic signs by template matching. Further, the method provides the provision for warning the driver by use of the recognized data of the traffic signs.

Claims

1. A method for detecting and recognizing traffic signs using images captured by a digital color and night vision camera, the said method characterized in being illumination invariant comprising:

transforming, by a processor, RGB image into HSV color model and subsequently extracting desired color components by using color quantization;

filtering, by the processor, noise components in the HSV color model in order to obtain a filtered image, wherein the filtered image is obtained by,

computing a centroid of each Region of Interest (ROI) in the HSV color model,

calculating a left distance indicating a distance from the centroid to a plurality of left side elements present in the HSV color model,

calculating a right distance indicating a distance from the centroid to a plurality of right side elements present in the HSV color model,

filtering the noise components based on the left distance, the right distance, a count of the plurality of left side elements, and a count of the plurality of right side elements;

detecting, by the processor, edges of objects in the filtered image and subsequently detecting distinct objects in the filtered image;

classifying, by the processor, shapes of traffic signs based on shape of the distinct objects; and

recognizing, by the processor, the shapes of the traffic signs by template matching.

2. The method of claim 1, wherein the shapes of the traffic signs comprises triangle, inverted triangle, rectangle, octagon, circle or ellipse.

3. The method of claim 1, wherein the shapes of the traffic signs comprise

distortions due to the noise components in the images captured by the digital color and the night vision camera;

shape distortions, in the real word traffic signs, due to problems in the manufacture of the traffic signs;

or

shape distortions due to different viewing distance.

4. The method of claim 1, wherein the shapes are classified by:

detecting lines in the shapes by using probabilistic hough transform;

determining line segments and subsequently merging the line segments if the line segments belong to lines; and

calculating sum of interior angles defined by the endpoints for differentiating triangle, inverted triangle, rectangle or octagon shapes of traffic signs.

5. The method of claim 1, wherein the shapes are classified by:

detecting lines in the shapes by using probabilistic hough transform;

determining line segments and subsequently merging the line segments if the line segments belong to lines; and

detecting major and minor axes length of a traffic sign for differentiating a circle or an ellipse shapes of the traffic signs.

6. The method of claim 1, further comprising the provision for warning the driver based on the shapes of traffic signs.

7. The method of claim 1, wherein the digital color and the night vision camera are disposed either on a dashboard or in front or inside or top of a vehicle.

11. The method of claim 1, wherein the filtered image includes at least one object, present in the HSV color model, which is in accordance with a pre-defined equation, and wherein the pre-defined equation is:

sum(left distance/right distance)≈

max(count of the plurality of left side elements and count of the plurality of right side elements).

8. An apparatus for detecting and recognizing traffic signs, the apparatus characterized in being illumination invariant, comprising of:

a digital color and a night vision camera disposed on the vehicle for capturing an image; and

a processor, coupled with a memory, embedded therein for analyzing the image in real-time for detecting and recognizing the traffic signs by;

transforming RGB image into HSV color model and subsequently extracting desired color components by using color quantization;

filtering noise components in the HSV color model in order to obtain a filtered image, wherein the filtered image is obtained by;

computing a centroid of each Region of Interest (ROI) in the HSV color model,

calculating a left distance indicating a distance from the centroid to a plurality of left side elements present in the HSV color model,

calculating a right distance indicating a distance from the centroid to a plurality of right side elements present in the HSV color model,

filtering the noise components based on the left distance, the right distance, a count of the plurality of left side elements, and a count of the plurality of right side elements;

detecting edges of objects in the filtered image and subsequently detecting distinct objects in the filtered image;

classifying shapes of traffic signs based on shape of the distinct objects; and

recognizing the shapes of the traffic signs by template matching.

9. The apparatus of claim 8, further comprises an alert means for warning the driver based on the shapes of the traffic signs.

10. The apparatus of claim 8, wherein the digital color and night vision camera are disposed either on a dashboard or in front or inside or top of a vehicle.