Back to results
US 7,610,250 B2Granted
Real-time method of determining eye closure state using off-line adaboost-over-genetic programming
AssigneeDelphi Technologies, Inc.
InventorsBranislav Kisacanin, Eric Yoder
Life of this patent
Estimating the term…
Abstract
Real-time video images of a human subject's face are processed by a plurality of classification algorithms developed in an off-line training process to determine the open vs. closed eye state of the subject. The off-line training process utilizes a genetic programming loop embedded within an adaptive boosting loop, and forms the classification algorithms and weighting factors for combining their classification scores. In the real-time process, the individual classification scores are combined and compared to a threshold to determine the open vs. closed eye state.
Claims4
A method of determining an open vs. closed state of an eye of a human subject, the method comprising the steps of:
- (a) forming training examples based on training images of human eyes, each training example including a feature vector characterizing a respective training image and a desired open vs. closed state of the human eye in that training image;
- (b) producing a prescribed number of classification programs, each of which is configured to evaluate a feature vector characterizing an image of a human eye and generate an eye state output corresponding to an open vs. closed state of the imaged eye;
- (c) performing a genetic programming process in which the produced classification programs are genetically modified over a prescribed number of generations;
- (d) applying the training examples to each of the genetically modified classification programs to produce a plurality of eye state outputs, determining deviations of the produced eye state outputs from the desired open vs. closed eye states of the training examples, and accumulating the determined deviations to form a classification error for each of the genetically modified classification programs;
- (e) selecting and saving the genetically modified classification program whose classification error is smallest;
- (f) repeating steps (c), (d) and (e) one or more times to form a set of saved genetically modified classification programs;
- (g) storing the set of saved genetically modified classification programs in a vision-based monitoring system including a microprocessor-based controller;
- (h) using the vision-based monitoring system to obtain a real-time video image of the human subject's eye and processing said real-time video image with said controller to generate a non-training feature vector characterizing said video image;
- (i) applying the non-training feature vector to each of the genetically modified classification programs stored in the vision-based monitoring system and collecting the eye state outputs generated by the classification programs;
- (j) combining the collected eye state outputs to determine an open vs. closed state of said human subject's eye;
- and
- (k) outputting the determined open vs. closed state from said controller.
9 more elements
A method of determining an open vs. closed state of an eye of a human subject, the method comprising the steps of:
- (a) using a vision-based monitoring system to obtain a real-time video image of the human subject's eye and processing said real-time video image with a microprocessor-based controller to generate a feature vector characterizing said real-time video image;
- (b) individually applying the feature vector to each of a plurality of genetically modified classification programs stored in said microprocessor-based controller to produce a plurality of open vs. closed eye state outputs, where the stored classification programs are developed by:
- (i) forming training examples based on training images of human eyes, each training example including a feature vector characterizing a respective training image and a desired open vs. closed state of the human eye in that training image;
- (ii) producing a prescribed number of classification programs, each of which is configured to evaluate a feature vector characterizing an image of a human eye and generate an eye state output corresponding to an open vs. closed state of the imaged eye;
- (iii) performing a genetic programming process in which the produced classification programs are genetically modified over a prescribed number of generations;
- (iv) applying the training examples to each of the genetically modified classification programs, and determining a classification error for each genetically modified classification program based on a deviation of its eye state outputs from the desired open vs. closed eye states of the training examples;
- (v) selecting and saving the genetically modified classification program whose classification error is smallest;
- (vi) repeating steps (iii), (iv) and (v) one or more times to form a set of saved genetically modified classification programs;
- (vii) storing the saved genetically modified classification programs in said microprocessor-based controller, and
- (c) combining said plurality of open vs. closed eye state outputs to determine the open vs. closed state of the human subject's eye;
- and
- (d) outputting the determined open vs. closed state from said controller.
9 more elements
Dependencies are inferred from the claim text.