Dynamic Bayesian Networks for vehicle classification in video
First Claim
Patent Images
1. A method for vehicle classification comprising:
- detecting at least three subcomponents, the at least three subcomponents comprising vehicle detection, license plate extraction, and tail light extraction;
using a Gaussian mixture model approach for detection of a moving object, wherein the Gaussian mixture model comprises Gaussian distributions to determine if a pixel is more likely to belong to a background model or not, and an AND approach, which determines a pixel as background only if the pixel falls within three standard deviations for all the components in all three R, G, and B color channels;
validating detected moving objects by using a simple frame differencing approach;
removing shadows and erroneous pixels by finding a vertical axis of symmetry using an accelerated version of Loy'"'"'s symmetry and readjusting a bounding box containing a mask with respect to an axis of symmetry, wherein if the shadow is behind the vehicle, removing the shadow using geometrical assumptions such as camera location, object geometry, and ground surface geometry, and wherein given the vehicle rear mask, measuring a height and width of the bounding box, and area of the mask;
inputting the license plate corner coordinates into an algorithm for license plate extraction, and adding a Gaussian noise with constant mean 0 and variance 0.2 times width to the license plate width measurement; and
performing a Bayesian network analysis on the at least three detected subcomponents for a plurality of vehicle classifications, wherein the Bayesian network analysis is defined as a directed acyclic graph G=(V, E), where nodes represent random variables from a domain of interest and arcs symbolize direct dependencies between the random variables.
0 Assignments
0 Petitions
Accused Products
Abstract
A system and method for classification of passenger vehicles and measuring their properties, and more particularly to a stochastic multi-class vehicle classification system, which classifies a vehicle (given its direct rear-side view) into one of four classes Sedan, Pickup truck, SUV/Minivan, and unknown, and wherein a feature pool of tail light and vehicle dimensions is extracted which feeds a feature selection algorithm to define a low-dimensional feature vector, and the feature vector is then processed by a Hybrid Dynamic Bayesian Network (HDBN) to classify each vehicle.
-
Citations
32 Claims
-
1. A method for vehicle classification comprising:
-
detecting at least three subcomponents, the at least three subcomponents comprising vehicle detection, license plate extraction, and tail light extraction; using a Gaussian mixture model approach for detection of a moving object, wherein the Gaussian mixture model comprises Gaussian distributions to determine if a pixel is more likely to belong to a background model or not, and an AND approach, which determines a pixel as background only if the pixel falls within three standard deviations for all the components in all three R, G, and B color channels; validating detected moving objects by using a simple frame differencing approach; removing shadows and erroneous pixels by finding a vertical axis of symmetry using an accelerated version of Loy'"'"'s symmetry and readjusting a bounding box containing a mask with respect to an axis of symmetry, wherein if the shadow is behind the vehicle, removing the shadow using geometrical assumptions such as camera location, object geometry, and ground surface geometry, and wherein given the vehicle rear mask, measuring a height and width of the bounding box, and area of the mask; inputting the license plate corner coordinates into an algorithm for license plate extraction, and adding a Gaussian noise with constant mean 0 and variance 0.2 times width to the license plate width measurement; and performing a Bayesian network analysis on the at least three detected subcomponents for a plurality of vehicle classifications, wherein the Bayesian network analysis is defined as a directed acyclic graph G=(V, E), where nodes represent random variables from a domain of interest and arcs symbolize direct dependencies between the random variables. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23)
-
-
24. A system for classification of vehicles comprising:
-
a camera configured to capture images of at least one moving object; and a computer processing unit configured to detecting at least three subcomponents from the captured images of the at least one moving object, the at least three subcomponents comprising vehicle detection, license plate extraction, and tail light extraction; use a Gaussian mixture model approach for detection of a moving object, wherein the Gaussian mixture model comprises Gaussian distributions to determine if a pixel is more likely to belong to a background model or not, and an AND approach, which determines a pixel as background only if the pixel falls within three standard deviations for all the components in all three R, G, and B color channels; validating detected moving objects by using a simple frame differencinq approach; remove shadows and erroneous pixels by finding a vertical axis of symmetry using an accelerated version of Loy'"'"'s symmetry and readiusting a bounding box containing a mask with respect to an axis of symmetry, wherein if the shadow is behind the vehicle, removing the shadow using geometrical assumptions such as camera location, object geometry, and ground surface geometry, and wherein given the vehicle rear mask, measuring a height and width of the bounding box, and area of the mask; input the license plate corner coordinates into an algorithm for license plate extraction, and adding a Gaussian noise with constant mean 0 and variance 0.2 times width to the license plate width measurement; and perform a Bayesian network analysis on the at least three detected subcomponents for a plurality of vehicle classifications, wherein the Bayesian network is defined as a directed acyclic graph G=(V, E), where nodes represent random variables from a domain of interest and arcs symbolize direct dependencies between random variables. - View Dependent Claims (25, 26, 28, 29, 30, 31, 32)
-
-
27. A computer program product comprising a non-transitory computer usable medium having a computer readable code embodied therein for classification of passenger vehicles and measuring their properties from a rear view video frame, the computer readable program code is configured to execute a process, which:
-
detects at least three subcomponents from the captured images of the at least one moving object, the at least three subcomponents comprising vehicle detection, license plate extraction, and tail light extraction; uses a Gaussian mixture model approach for detection of a moving object, wherein the Gaussian mixture model comprises Gaussian distributions to determine if a pixel is more likely to belong to a background model or not, and an AND approach, which determines a pixel as background only if the pixel falls within three standard deviations for all the components in all three R, G, and B color channels; validates detected moving objects by using a simple frame differencing approach; removes shadows and erroneous pixels by finding a vertical axis of symmetry using an accelerated version of Loy'"'"'s symmetry and readjusting a bounding box containing a mask with respect to an axis of symmetry, wherein if the shadow is behind the vehicle, removing the shadow using geometrical assumptions such as camera location, object geometry, and ground surface geometry, and wherein given the vehicle rear mask, measuring a height and width of the bounding box, and area of the mask; inputs the license plate corner coordinates into an algorithm for license plate extraction, and adding a Gaussian noise with constant mean 0 and variance 0.2 times width to the license plate width measurement; and performs a Bayesian network analysis on the at least three detected subcomponents for a plurality of vehicle classifications, which are known, wherein the Bayesian network is defined as a directed acyclic graph G=(V, E), where nodes represent random variables from a domain of interest and arcs symbolize direct dependencies between random variables.
-
Specification