Abstract: A system and method for finding edge orientation and magnitude, and extracting a road path from an image. An image is segmented into at least a first window and a second window. A first road segment having plural parameters within the first window is specified. Plural road segments are identified, each having plural parameters, in the second window. The orientation of plural line segments in the second window are determined, and paired, based on their orientation. The line segment orientations are determined by receiving a plurality of edge orientation estimates, transforming a portion of them, and aggregating the transformed portion into an edge orientation estimate. A stochastic process is performed on the plural parameters of the plural road segments to identify a second road segment having the maximum correlation with the plural parameters of the first road segment.

Abstract: A method is provided for tracking a non-rigid object in a sequence of frames of a video. Features of an object are extracted from the video. The features include locations of pixels and properties of the pixels. The features are used to construct a covariance matrix. The covariance matrix is used as a descriptor of the object for tracking purposes. Object deformations and appearance changes are managed with an update mechanism that is based on Lie algebra averaging.

Abstract: A method constructs descriptors for a set of data samples and determines a distance score between pairs of subsets selected from the set of data samples. A d-dimensional feature vector is extracted for each sample in each subset of samples. The feature vector includes indices to the corresponding sample and properties of the sample. The feature vectors of each subset of samples are combined into a d×d dimensional covariance matrix. The covariance matrix is a descriptor of the corresponding subset of samples. Then, a distance score is determined between the two subsets of samples using the descriptors to measure a similarity between the descriptors.

Abstract: A method models a scene. A video is acquired of the scene, and for each frame of the video, the method updates a set of background models for each pixel; a set of shadow models for each pixel; a set of shadow flow vectors for each color; and a background shadow map. Each pixel in each background model and each shadow model is represented by multiple layers. Each layer includes Gaussian distributions and each Gaussian distribution includes a mean and a covariance. The covariance is an inverse Wishart distribution and the updating is according to a recursive Bayesian estimation process.

Abstract: A method tracks non-rigid objects in a video acquired of a cluttered scene by a camera. The method uses a particle filter. The tracking includes the following steps: motion transition estimation, contour deformation detection, and contour regulation. The method uses a dynamic affine transform model and employs the particle filter to estimate the parameters of the model. The method generates a probabilistic map of deformation for tracking the contour of the object followed by a projection step to constrain or regulate the contour in a contour subspace.

Abstract: A video is acquired of a scene. Each pixel in each frame of the video is represented by multiple of layers. Each layer includes multiple Gaussian distributions. Each Gaussian distribution includes a mean and a covariance. The covariance is an inverse Wishart distribution. Then, the layers are updated for each frame with a recursive Bayesian estimation process to construct a model of the scene. The model can be used to detect foreground and background pixels according to confidence scores of the layers.

Abstract: A method tracks a moving object in a video acquired of a scene with a camera. A background model is maintained for each frame, and moving objects are detected according to changes in the background model. An object model is maintained for the moving object, and kernels are generated for the moving object. A mean-shift process is applied to each kernel in each frame to determine a likelihood of an estimated location of the moving object in each frame, according to the background models, the object model, and the mean shift kernels to track the moving object in the video.

Abstract: A computer implemented method extracts an integral histogram from sampled data, such as time series data, images, and volumetric data. First, a set of samples is acquired from a real-word signal. The set of samples is scanned in a predetermined order. For each current sample, an integral histogram integrating a histogram of the current sample and integral histograms of previously scanned samples is constructed.

Abstract: A computer implemented method models a scene in a video acquired by a camera. For each pixel in each frame of the video, a time series of intensities of the pixel is acquired. A harmonic series is extracted from samples of each time series using a sliding window. Distances between the harmonic series are measured. The distances are an estimate of spectral components in an autocorrelation function of underlying dynamic processes in the scene.

Abstract: A temporal sequence of images is acquired of a dynamic scene. Spatial gradients are determined using filters. By taking advantage of the sparseness of outputs of the filters, an intrinsic background image is generated as median filtered gradients. The intrinsic background image is then divided into the original sequence of images to yield intrinsic foreground images. The intrinsic foreground images can be thresholded to obtain a detection mask.

Abstract: A method detects roads in an aerial image of ground topology by determining low-level features, such as intensities and gradients, for each pixel in the aerial image, determining middle-level features, such as an orientation for each pixel from the low-level features, and determining high-level features from the middle-level features. Each high-level feature is assigned a probability, and the probabilities of the high-level features for each pixel are normalized and aggregated to a single probability that the pixel is associated with a road.

Abstract: A method determines a warping function between a first image and a second image. For each pixel in each scan-line of the first image, a distance matrix is constructed which represents distances between the pixel and a plurality of corresponding pixels of a second image. The distance matrices are ordered according a scan-line order of the pixels.

Abstract: A method models a set of trajectories of a set of objects; there is one trajectory for each object. A sequence of positions of each object is obtained over a duration of time to define a trajectory of the object. The sequence of positions is then projected to a hidden Markov model to model the trajectory of each object.

Abstract: The invention provides a method for detecting usual events in a video. The events are detected by first constructing an aggregate affinity matrix from features of associated items extracted from the video. The affinity matrix is decomposed into eigenvectors, and the eigenvectors are used to reconstruct approximate estimates of the aggregate affinity matrix. Each matrix is clustered and scored, and the clustering that yields the highest scores is used to detect usual events.

Abstract: The invention provides a method for detecting usual and unusual events in a video. The events are detected by first constructing an aggregate affinity matrix from features of associated items extracted from the video. The affinity matrix is decomposed into eigenvectors, and the eigenvectors are used to reconstruct approximate estimates of the aggregate affinity matrix. Each matrix is clustered and scored, and the clustering that yields the highest scores is used to detect events.

Abstract: A method partitions an image into segments. The image includes pixel points. A set of base points is selected in the image. A wavefront is initialized and emitted from each base point. Each wavefront is propagated according to a speed function until a termination condition is satisfied to determine a corresponding final wavefront. The image is then segmented according to each final wavefront.

Abstract: A method detects traffic events in a compressed video. Feature vectors are extracted from the compressed video. The feature vector are provided to a Gaussian mixture hidden Markov model. Then, a maximum likelihood of the Gaussian mixture hidden Markov model is determined to classify the plurality of feature vector as traffic events.

Abstract: A method compares a first data sequence with a second data sequence. A cross-correlation matrix of distances between all possible pairs of values of the first and second data sequences is constructed. A minimum cost path between diagonally opposing corners of the cross-correlation matrix is determined. The minimum cost path is projected onto a transfer function that maps between the first data sequence and the second data sequence. Then, the minimum cost path is projected onto a deformation function that corresponds to a similarity between the first data sequence and the second data sequence.

Abstract: A system and method for finding edge orientation and magnitude, and extracting a road path from an image. An image is segmented into at least a first window and a second window. A first road segment having plural parameters within the first window is specified. Plural road segments are identified, each having plural parameters, in the second window. The orientation of plural line segments in the second window are determined, and paired, based on their orientation. The line segment orientations are determined by receiving a plurality of edge orientation estimates, transforming a portion of them, and aggregating the transformed portion into an edge orientation estimate. A stochastic process is performed on the plural parameters of the plural road segments to identify a second road segment having the maximum correlation with the plural parameters of the first road segment.