Systems and Methods for Identifying Traffic Control Devices and Testing the Retroreflectivity of the Same

Patent metadata

Field	Value
Publication number	US-2017193312-A1
Application number	US-201515129655-A
Country	US
Kind code	A1
Filing date	Mar 27, 2015
Priority date	Mar 27, 2014
Publication date	Jul 6, 2017
Grant date	—

Abstract

Systems and methods for identifying traffic control devices from images, and systems and methods for assessing the retro reflectivity of traffic control devices. The identification of traffic control devices can be accomplished using a lighting-dependent statistical color model. The identification of traffic control devices can be accomplished using an active contour or active polygon method.

First claim

1 . A method for identifying a traffic sign comprising:
classifying an image as having a lighting condition; segmenting the image into a color used for traffic signs using a statistical color model specific to the lighting condition; and detecting a shape in the image corresponding to the traffic sign.

2 . The method of claim 1 , wherein the color used for traffic signs is a MUTCD standard color.

3 . The method of claim 1 , wherein classifying the image further comprises classifying the image as underexposed where a mean pixel brightness value of the image is below an under-saturation threshold.

4 . The method of claim 1 , wherein classifying the image further comprises classifying the image as overexposed where a mean pixel brightness value of the image is above an over-saturation threshold.

5 . The method of claim 1 , wherein classifying the image further comprises classifying the image as adverse lighting if the difference between a mean pixel brightness of the image and a median pixel brightness of the image is over an adverse lighting threshold.

6 . The method of claim 1 , wherein classifying an image further comprises classifying the image as normal if:
a mean pixel brightness value of the image is above an under-saturation threshold; a mean pixel brightness value of the image is below an over-saturation threshold; and a difference between a mean pixel brightness of the image and median pixel brightness of the image is below an adverse lighting threshold.

7 . The method of claim 5 , wherein images having a lighting condition of adverse lighting are divided into a region having an over-exposed condition, and a region having an under-exposed condition, by:
generating a threshold surface; comparing the threshold surface to the image to create a thresholded image; identifying candidate regions of the image; and applying a morphological open and close operation to the candidate regions of the image.

8 . The method of claim 7 , wherein generating a threshold surface is accomplished using an anti-geometric heat equation.

9 . The method of claim 1 , wherein segmenting an image further comprises:
calculating, for a plurality of pixels, a local pixel-level homogeneity value for one of a hue, saturation, and value; normalizing the local pixel-level homogenity value; and generating a probability distribution by applying an artificial neural network specific to a lighting condition, having input values of hue, saturation, value, and one or more of hue homogeneity, saturation homogeneity, and value homogeneity.

10 . The method of claim 9 , wherein the artificial neural network is a functional link network.

11 . The method of claim 1 , wherein the detecting step is performed by an differential equation based shape detection algorithm.

12 . The method of claim 11 , wherein the differential equation based shape detection algorithm comprises a region-based energy function.

13 . The method of claim 11 , wherein the differential equation based shape detection algorithm comprises an active contour algorithm.

14 . The method of claim 13 , wherein the active contour function comprises a probability distribution function sub-energy component that represents the probability of a sign image occurring in each pixel.

15 . The method of claim 13 , wherein the active contour function comprises a statistical color model sub-energy component represents the probability of a traffic sign color occurring in each pixel of the image.

16 . The method of claim 13 , wherein the active contour function comprises a global contour length sub-energy component with a maximum contour length.

17 . The method of claim 16 , wherein the maximum contour length is calculated as a function of a total perimeter of the image.

18 . The method of claim 11 , wherein the differential equation based shape detection algorithm comprises an active polygon algorithm.

19 . The method of claim 18 , wherein the active polygon contour algorithm comprises a generalized Hough transform.

20 . The method of claim 19 , wherein the generalized Hough transform comprises:
calculating an R-table corresponding to the shape of a traffic sign; detecting the center where the maximum similarity is obtained compared to the R-table; and solving the region-based energy function for the optimal value.

21 . A method of assessing the retroreflectivity condition of a traffic sign comprising:
receiving, at a processor and from a LiDAR sensor, a plurality of LiDAR data points, each LiDAR data point in the plurality of LiDAR data points relating to a location on the face of the traffic sign, each LiDAR data point comprising 3D position information and a set of retro-intensity data, wherein each set of retro-intensity data comprises:
a retro-intensity value;
a distance value; and
an angle value;
determining, for each LiDAR data point, an incidence angle value; receiving a plurality of image data points, wherein each image data point represents a portion of a traffic sign image, each image data point comprising:
color data; and
2D location data representing a location on the face of the traffic sign;
associating each of a plurality of LiDAR data points with a corresponding image data point, wherein 2D location data of a particular image data point corresponds to a location on the face of the traffic sign from which a particular LiDAR data point associated with the particular image data point relates; grouping each LiDAR data point into one or more color clusters based on the associated color data, normalizing, for each color cluster of LiDAR data points, each retro-intensity value based on the corresponding distance value and incidence angle value; and determining, for each color cluster of LiDAR data points, whether the normalized retro-intensity values indicate a retroreflectivity above a predetermined threshold.

22 . The method of claim 21 , wherein the 3D position information comprises latitude data, longitude data and elevation data.

23 . The method of claim 21 , wherein each retro-intensity value represents a ratio of energy redirected from the traffic sign to the energy emitted from the LiDAR sensor.

24 . The method of claim 21 , wherein the distance value is a value that is representative of the distance between the traffic sign and the LiDAR sensor at the time of the measurement of the LiDAR data point.

25 . The method of claim 21 , wherein the angle value represents a LiDAR beam angle with respect to the level of the LiDAR sensor.

26 . The method of claim 21 , wherein the portion of the traffic sign image comprises a pixel.

27 . The method of claim 21 , wherein the color data represents the color of the portion of the traffic sign image.

28 . The method of claim 21 , wherein the 2D location data represents the location of the portion of the traffic sign image on a face of the traffic sign.

29 . The method of claim 21 , wherein determining whether the normalized retro-intensity values indicate a retroreflectivity above a predetermined threshold based on the color comprises:
determining a median value the normalized retro-intensity values for a color cluster of LiDAR data points; and comparing the median value to a predetermined threshold associated with the color of the color cluster of the median value.

3

Assignees

• Georgia Tech Res Inst

Inventors

• Ai Chengbo
• Tsai Yichang
• Wang Zhaohua

Classifications

• G09F2009/3055

  for traffic signs · CPC title

• G01S17/42

  Simultaneous measurement of distance and other co-ordinates (indirect measurement G01S17/46) · CPC title

• G06F13/20

  for access to input/output bus · CPC title

• G08G1/00

  Traffic control systems for road vehicles (arrangement of road signs or traffic signals E01F9/00 {; automatic vehicle control B62D}) · CPC title

• G01S17/931

  of land vehicles · CPC title