High-accuracy real-time road sign detection from images
First Claim
Patent Images
1. A method of detecting objects, comprising:
- receiving images from an image capture device;
detecting a plurality of candidate objects within the received images, comprising;
performing constant-time normalized cross-correlation between a template and the camera image, the constant-time normalized cross-correlation comprising detecting areas of a given color within the received image by computing an intermediate image;
building a first integral image from the intermediate image; and
computing a second integral image which sums the squares of the intermediate image;
computing a score for objects identified in the camera image based on their fit within the template; and
selecting the candidate objects based on the computed score;
performing shape classification on the plurality of candidate objects; and
eliminating non-objects from the plurality of candidate objects.
3 Assignments
0 Petitions
Accused Products
Abstract
Objects, such as road signs, may be detected in real-time using a camera or other image capture device. As images are received through the camera, candidate signs are first detected. The detection of candidate signs employs constant-time normalized cross correlation, including generation of intermediate images and integral images, and applying a template of concentric, different sized shapes over the integral images. From the pool of candidate signs, false positives may be separated out using shape classification to identify actual road signs.
50 Citations
17 Claims
-
1. A method of detecting objects, comprising:
-
receiving images from an image capture device; detecting a plurality of candidate objects within the received images, comprising; performing constant-time normalized cross-correlation between a template and the camera image, the constant-time normalized cross-correlation comprising detecting areas of a given color within the received image by computing an intermediate image;
building a first integral image from the intermediate image; and
computing a second integral image which sums the squares of the intermediate image;computing a score for objects identified in the camera image based on their fit within the template; and selecting the candidate objects based on the computed score; performing shape classification on the plurality of candidate objects; and eliminating non-objects from the plurality of candidate objects. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9)
-
-
10. A system for detecting objects, comprising:
-
an image capture device adapted to continually receive images; a processor in communication with the image capture device; a memory storing instructions executable by the processor, the instructions for performing a method of; receiving images from the image capture device; detecting a plurality of candidate objects within the received images, comprising; performing constant-time normalized cross-correlation between a template and the camera image, the constant-time normalized cross-correlation comprising detecting areas of a given color within the received image by computing an intermediate image;
building a first integral image from the intermediate image; and
computing a second integral image which sums the squares of the intermediate image;computing a score for objects identified in the camera image based on their fit within the template; and selecting the candidate objects based on the computed score; performing shape classification on the plurality of candidate objects; and eliminating non-objects from the plurality of candidate objects. - View Dependent Claims (11, 12, 13, 14, 15, 16)
-
-
17. A non-transitory computer readable medium storing information and instructions executable by a processor for performing a method of detecting objects, comprising:
-
receiving images from an image capture device; detecting a plurality of candidate objects within the received images, comprising; performing constant-time normalized cross-correlation between a template and the camera image, the constant-time normalized cross-correlation comprising detecting areas of a given color within the received image by computing an intermediate image;
building a first integral image from the intermediate image; and
computing a second integral image which sums the squares of the intermediate image;computing a score for objects identified in the camera image based on their fit within the template; and selecting the candidate objects based on the computed score; performing shape classification on the plurality of candidate objects; and eliminating non-objects from the plurality of candidate objects.
-
Specification