Abstract: The present disclosure relates generally to optical flow algorithms. Section 1 of the present disclosure describes an optical flow algorithm with real-time performance and adequate accuracy for embedded vision applications. This optical flow algorithm is based on a ridge estimator. Sections 2 and 3 describe an obstacle detection algorithm that utilizes the motion field that is output from the optical flow algorithm. Section 2 is focused on unmanned ground vehicles, whereas section 3 is focused on unmanned aerial vehicles.

Abstract: A method for stereo vision may include filtering a row or column in a stereo image to obtain intensity profiles, identifying peaks in the intensity profiles, pairing peaks within a maximum disparity distance, determining a shape interval for peak pairs, selecting a peak pair with a maximum shape interval, determining a disparity offset for the peak pairs, extending shape intervals to include all pixels in the intensity profiles, computing depths or distances from disparity offsets, and smoothing the stereo image disparity map along a perpendicular dimension. Another method for stereo vision includes filtering stereo images to intensity profiles, identifying peaks in the intensity profiles, pairing peaks within a maximum disparity distance, determining shape intervals for peak pairs, and selecting peak pairs with the maximum shape interval. Apparatus corresponding to the above methods are also disclosed herein.

Abstract: A method for training an image processing neural network without human selection of features may include providing a training set of images labeled with two or more classifications, providing an image processing toolbox with image transforms that can be applied to the training set, generating a random set of feature extraction pipelines, where each feature extraction pipeline includes a sequence of image transforms randomly selected from the image processing toolbox and randomly selected control parameters associated with the sequence of image transforms. The method may also include coupling a first stage classifier to an output of each feature extraction pipeline and executing a genetic algorithm to conduct genetic modification of each feature extraction pipeline and train each first stage classifier on the training set, and coupling a second stage classifier to each of the first stage classifiers in order to increase classification accuracy.

Abstract: A method for stereo vision may include filtering a row or column in a stereo image to obtain intensity profiles, identifying peaks in the intensity profiles, pairing peaks within a maximum disparity distance, determining a shape interval for peak pairs, selecting a peak pair with a maximum shape interval, determining a disparity offset for the peak pairs, extending shape intervals to include all pixels in the intensity profiles, computing depths or distances from disparity offsets, and smoothing the stereo image disparity map along a perpendicular dimension. Another method for stereo vision includes filtering stereo images to intensity profiles, identifying peaks in the intensity profiles, pairing peaks within a maximum disparity distance, determining shape intervals for peak pairs, and selecting peak pairs with the maximum shape interval. Apparatus corresponding to the above methods are also disclosed herein.

Abstract: A method for training an image processing neural network without human selection of features may include providing a training set of images labeled with two or more classifications, providing an image processing toolbox with image transforms that can be applied to the training set, generating a random set of feature extraction pipelines, where each feature extraction pipeline includes a sequence of image transforms randomly selected from the image processing toolbox and randomly selected control parameters associated with the sequence of image transforms. The method may also include coupling a first stage classifier to an output of each feature extraction pipeline and executing a genetic algorithm to conduct genetic modification of each feature extraction pipeline and train each first stage classifier on the training set, and coupling a second stage classifier to each of the first stage classifiers in order to increase classification accuracy.

Abstract: The present disclosure relates generally to optical flow algorithms. Section 1 of the present disclosure describes an optical flow algorithm with real-time performance and adequate accuracy for embedded vision applications. This optical flow algorithm is based on a ridge estimator. Sections 2 and 3 describe an obstacle detection algorithm that utilizes the motion field that is output from the optical flow algorithm. Section 2 is focused on unmanned ground vehicles, whereas section 3 is focused on unmanned aerial vehicles.

Abstract: The provision of speech therapy to a learner (76) entails receiving a speech signal (156) from the learner (76) at a computing system (24). The speech signal (156) corresponds to an utterance (116) made by the learner (76). A set of parameters (166) is ascertained from the speech signal (156). The parameters (166) represent a contact pattern (52) between a tongue and palate of the learner (156) during the utterance (116). For each parameter in the set of parameters (166), a deviation measure (188) is calculated relative to a corresponding parameter from a set of normative parameters (138) characterizing an ideal pronunciation of the utterance (116). An accuracy score (56) for the utterance (116), relative to its ideal pronunciation, is generated from the deviation measure (188). The accuracy score (56) is provided to the learner (76) to visualize accuracy of the utterance (116) relative to its ideal pronunciation.

Abstract: A three-dimensional measurement system and method for objects, such as oysters, projects one or more laser lines onto a surface on which the object is currently located. The laser lines are picked up as parallel lines by a camera where no object is located on the surface. When an object is located on the surface, the camera obtains an image that includes lines displaced from the parallel lines as a result of the lines impinging on portions of the object that have a particular height associated therewith. The displacement data allows a processor to determine the height of the object at various positions of the object can be obtained. Volume can be obtained using a binary image of the object to calculate area from the height data. The object can then be classified according to its volume.