Abstract: Described herein are systems and methods for multimodal recurrent network processing. In an embodiment, a system for evaluating multimodal data comprising a multimodal data input and a multimodal processing module is described. The multimodal data input may comprise the multimodal data, the multimodal data may comprise a first modality and a second modality. The multimodal processing module may be configured to receive the multimodal data comprising the first modality and the second modality; evaluate the first modality using a first recursive neural network comprising a first transformation matrix; evaluate the second modality using a second recursive neural network comprising the first transformation matrix; and determine an output based, at least in part, on evaluating the first modality and the second modality.

Abstract: Described herein are systems and methods for multimodal recurrent network processing. In an embodiment, a system for evaluating multimodal data comprising a multimodal data input and a multimodal processing module is described. The multimodal data input may comprise the multimodal data, the multimodal data may comprise a first modality and a second modality. The multimodal processing module may be configured to receive the multimodal data comprising the first modality and the second modality; evaluate the first modality using a first recursive neural network comprising a first transformation matrix; evaluate the second modality using a second recursive neural network comprising the first transformation matrix; and determine an output based, at least in part, on evaluating the first modality and the second modality.

Abstract: In accordance with various embodiments of the disclosed subject matter, a method and a system for detecting multiple objects from real-time images are provided. The method comprises: performing, using a CPU host, an image segmentation process to divide real-time input images into a plurality of image partitions; performing, by multiple GPUs, a fast block-wise registration process, a mark setting process, a background generation process, a foreground generation process based on a Hyper-Q computation infrastructure, and a support vector machine classification process; and generating, by the CPU host, visualization classification images.

Abstract: In accordance with various embodiments of the disclosed subject matter, a method and a system for detecting multiple objects from real-time images are provided. The method comprises: performing, using a CPU host, an image segmentation process to divide real-time input images into a plurality of image partitions; performing, by multiple GPUs, a fast block-wise registration process, a mark setting process, a background generation process, a foreground generation process based on a Hyper-Q computation infrastructure, and a support vector machine classification process; and generating, by the CPU host, visualization classification images.

Abstract: Methods and devices for demonstrating three-player pursuit-evasion (PE) game are provided using a hardware-in-loop test-bed. Robots including pursuer robots and an evader robot are arranged on a solid surface. A drone is positioned flying above to oversee the robots to capture a video or an image sequence of the robots. A robot thread process and a drone thread process are implemented by a computer. In the robot thread process, a tracking-by-detection process is perform to provide a state of the robot including a location and a heading direction of the robot; a delay compensation is conducted; and a PE game is called to calculate a robot command. In the drone thread process, a drone control is calculated to make the drone follow an evader robot, the drone control is sent to the drone, and user commands are also checked.

Abstract: A method and apparatus for tracking an object across a plurality of sequential images, where certain of the images contain motion blur. A plurality of normal templates of a clear target object image and a plurality of blur templates of the target object are generated. In the next subsequent image frame, a plurality of bounding boxes are generated of potential object tracking positions about the target object location in the preceding image frame. For each bounding box image frame, a reconstruction error is generated that one bounding box has a maximum probability that it is the object tracking result in the subsequent image frame.

Abstract: A method and apparatus for tracking an object across a plurality of sequential images, where certain of the images contain motion blur. A plurality of normal templates of a clear target object image and a plurality of blur templates of the target object are generated. In the next subsequent image frame, a plurality of bounding boxes are generated of potential object tracking positions about the target object location in the preceding image frame. For each bounding box image frame, a reconstruction error is generated that one bounding box has a maximum probability that it is the object tracking result in the subsequent image frame.

Abstract: Methods and devices for demonstrating three-player pursuit-evasion (PE) game are provided using a hardware-in-loop test-bed. Robots including pursuer robots and an evader robot are arranged on a solid surface. A drone is positioned flying above to oversee the robots to capture a video or an image sequence of the robots. A robot thread process and a drone thread process are implemented by a computer. In the robot thread process, a tracking-by-detection process is perform to provide a state of the robot including a location and a heading direction of the robot; a delay compensation is conducted; and a PE game is called to calculate a robot command. In the drone thread process, a drone control is calculated to make the drone follow an evader robot, the drone control is sent to the drone, and user commands are also checked.

Abstract: A method and apparatus for detecting 3D anatomical objects in medical images using constrained marginal space learning (MSL) is disclosed. A constrained search range is determined for an input medical image volume based on training data. A first trained classifier is used to detect position candidates in the constrained search range. Position-orientation hypotheses are generated from the position candidates using orientation examples in the training data. A second trained classifier is used to detect position-orientation candidates from the position-orientation hypotheses. Similarity transformation hypotheses are generated from the position-orientation candidates based on scale examples in the training data. A third trained classifier is used to detect similarity transformation candidates from the similarity transformation hypotheses, and the similarity transformation candidates define the position, translation, and scale of the 3D anatomic object in the medical image volume.

Abstract: A new measure to compare histogram-based descriptors, a diffusion distance, is disclosed. The difference between two histograms is defined to be a temperature field. The relationship between histogram similarity and diffusion process is discussed and it is shown how the diffusion handles deformation as well as quantization effects. As a result, the diffusion distance is derived as the sum of dissimilarities over scales. Being a cross-bin histogram distance, the diffusion distance is robust to deformation, lighting change and noise in histogram-based local descriptors. In addition, it enjoys linear computational complexity which significantly improves previously proposed cross-bin distances with quadratic complexity or higher The proposed approach is tested on both shape recognition and interest point matching tasks using several multi-dimensional histogram-based descriptors including shape context, SIFT and spin images.

Abstract: A method and apparatus for detecting 3D anatomical objects in medical images using constrained marginal space learning (MSL) is disclosed. A constrained search range is determined for an input medical image volume based on training data. A first trained classifier is used to detect position candidates in the constrained search range. Position-orientation hypotheses are generated from the position candidates using orientation examples in the training data. A second trained classifier is used to detect position-orientation candidates from the position-orientation hypotheses. Similarity transformation hypotheses are generated from the position-orientation candidates based on scale examples in the training data. A third trained classifier is used to detect similarity transformation candidates from the similarity transformation hypotheses, and the similarity transformation candidates define the position, translation, and scale of the 3D anatomic object in the medical image volume.

Abstract: A system for separating text and drawings in a digital ink file (e.g., a handwritten digital ink file). A stroke analyzer classifies single strokes that have been input by a user as “text” or “unknown.” The stroke analyzer utilizes a trainable classifier, such as a support vector machine. A grouping component is provided that groups text strokes in an attempt to form text objects (e.g., words, characters, or letters). The grouping component also groups unknown strokes in an attempt to form objects (e.g., shapes, drawings, or even text). A trainable classifier, such as a support vector machine, evaluates the grouped strokes to determine if they are objects.

Abstract: A new measure to compare histogram-based descriptors, a diffusion distance, is disclosed. The difference between two histograms is defined to be a temperature field. The relationship between histogram similarity and diffusion process is discussed and it is shown how the diffusion handles deformation as well as quantization effects. As a result, the diffusion distance is derived as the sum of dissimilarities over scales. Being a cross-bin histogram distance, the diffusion distance is robust to deformation, lighting change and noise in histogram-based local descriptors. In addition, it enjoys linear computational complexity which significantly improves previously proposed cross-bin distances with quadratic complexity or higher The proposed approach is tested on both shape recognition and interest point matching tasks using several multi-dimensional histogram-based descriptors including shape context, SIFT and spin images.

Abstract: A system for separating text and drawings in a digital ink file (e.g., a handwritten digital ink file). A stroke analyzer classifies single strokes that have been input by a user as “text” or “unknown.” The stroke analyzer utilizes a trainable classifier, such as a support vector machine. A grouping component is provided that groups text strokes in an attempt to form text objects (e.g., words, characters, or letters). The grouping component also groups unknown strokes in an attempt to form objects (e.g., shapes, drawings, or even text). A trainable classifier, such as a support vector machine, evaluates the grouped strokes to determine if they are objects.