Abstract: In accordance with various embodiments of the disclosed subject matter, a method and a system for vision-centric deep-learning-based road situation analysis are provided. The method can include: receiving real-time traffic environment visual input from a camera; determining, using a ROLO engine, at least one initial region of interest from the real-time traffic environment visual input by using a CNN training method; verifying the at least one initial region of interest to determine if a detected object in the at least one initial region of interest is a candidate object to be tracked; using LSTMs to track the detected object based on the real-time traffic environment visual input, and predicting a future status of the detected object by using the CNN training method; and determining if a warning signal is to be presented to a driver of a vehicle based on the predicted future status of the detected object.

Abstract: A method and a non-transitory computer-readable medium having computer program for optimizing a system function-flow of a system capable of performing at least one function and comprising a plurality of components by utilizing application programming interface (API) profiling are provided. The method comprises receiving a request to perform the at least one function of the system from a user; obtaining at least one component required to realize the at least one function requested by the user from the plurality of the components; providing a plurality of APIs to the at least one obtained component to choose from; profiling each of the plurality of APIs to get a performance score of each API; based on the profiled APIs, optimizing the system function-flow of the system; determining corresponding APIs for the optimized system function-flow; and performing the at least one function requested by the user based on the determined APIs.

Abstract: A video retrieval method is provided. The method includes dividing video sequences stored in a video database into a plurality of video frames, extracting by a predetermined feature extracting algorithm a plurality of selected fingerprints from the plurality of video frames representing features having characteristic information specific to the plurality of video frames, and storing the plurality of extracted selected fingerprints in a fingerprint database. The method also includes receiving a query video sequence submitted by a user, extracting by the predetermined feature extracting algorithm at least one query fingerprint representing at least one query frame from the received query video sequence, and calculating similarity between the query fingerprint and the selected fingerprints from the fingerprint database to search for a match for the query fingerprint.

Abstract: The present invention provides a method for converting a low dynamic range (LDR) image to a high dynamic range (HDR) image, including obtaining an input LDR image; and in an HDR image database, selecting one or more HDR images that match the input LDR image as candidate images. Further, the candidate images are warped according to a spatial correspondence between the candidate images and the input LDR image. The input LDR image is decomposed to an illumination component and a texture component. The illumination component and the texture component are reconstructed respectively based on the warped candidate images. The reconstructed illumination component and the reconstructed texture component are combined to produce an output HDR image.

Abstract: A method is provided for personalized video recommendation based on user interests modeling. The method includes detecting a viewing activity of at least one user of a content-presentation device capable of presenting multiple programs in one or more channels, and representing user interests of the at least one user by using a topic model. The method also includes discovering the user interests from user viewing histories, and generating a personalized video list of personalized video contents. Further, the method includes recommending the personalized video contents to the at least one user; and delivering the recommended personalized video to the at least one user such that the personalized video contents are presented on the content-presentation device.

Abstract: An inter-app switching method on a mobile device is provided. The method includes receiving at least one input associated with mobile app usage from a user and determining whether a pre-defined trigger is enabled based on the at least one received input. The method also includes performing routine user operations on the mobile device when the pre-defined trigger is disabled. Further, the method includes discovering a current user intent by utilizing at least one user interaction, recommending at least one mobile app recommendation and access point including at least an entrance to a function and a type of the function (FUNC) to the user, receiving a selection by the user from the at least one recommended FUNC, and directly switching from a current app page to another app page without returning to any home screen on the mobile device when the pre-defined trigger is enabled.

Abstract: A method is provided for recommending a desired func sequence to a user. The method includes obtaining a user intention list including at least one user intention; separating the user intention into a plurality of tasks; and creating a task flow graph for the plurality of tasks based on user usage data. Each vertex in the task flow graph represents a task and indicating an importance of the task. The method also includes creating a func flow graph based on the user usage data and temporal sequences of the tasks and funcs, and each vertex in the func flow graph represents a func and indicating an importance of the func. Further, the method includes determining a desired func sequence to fulfill the user intention based on the user usage data, the task flow graph, and the func flow graph; and recommending the desired func sequence to the user.

Abstract: A method for balancing recognition accuracy and power consumption is provided. The method includes activating a plurality of sensors onboard a device for a first time period and calculating sensor readings by using a calculation function. The method also includes determining a plurality of sensor subsets of the plurality of sensors and calculating corresponding errors for the plurality of sensor subsets for the first time period. Further, the method includes calculating respective power consumption of running the sensor subsets of the plurality of sensors from the plurality of sensor subsets that produce the error below the user-defined error threshold and selecting a sensor subset with minimum power consumption from the sensor subsets as an optimal sensor subset. In addition, the method includes deactivating the plurality of sensors not in the optimal sensor subset and calculating and displaying sensor readings with the optimal sensor subset for a second time period.

Abstract: The present invention provides a face recognition method on a computing device, comprising: storing a plurality of training face images, each training face image corresponding to a face class; obtaining one or more face test samples; applying a representation model to represent the face test sample as a combination of the training face images and error terms, wherein a coefficient vector is corresponded to the training face images; estimating the coefficient vector and the error terms by solving a constrained optimization problem; computing a residual error for each face class, the residual error for a face class being an error between the face test sample and the face test sample's representation model represented by the training samples in the face class; classifying the face test sample by selecting the face class that yields the minimal residual error; and presenting the face class of the face test sample.

Abstract: A personalized channel recommendation method includes recording video programs of a plurality of channels, extracting image features of the recorded video programs of the plurality of channels, and storing the extracted image features of the recorded video programs of the plurality of channels in a feature database. The method also includes obtaining image feature data extracted from a video program being viewed by at least one user of a user device on a current channel, and detecting that the video program being watched by the user is an undesired program and determining a channel number of a channel of the video program being watched based on the image feature data and the feature database. Further, the method includes determining channel recommendation including at least one desired channel to be switched to from the undesired program, and generating an adaptive TV guide for the user to skip the undesired program and to switch to the desired channel.

Abstract: A gesture based power management method for a wearable device is provided. The method includes establishing a mapping relationship between a set of pre-defined gestures including at least a lookup gesture and a set of power management functions including at least a function for turning on a display screen of the wearable device. The method also includes receiving motion information from a plurality of sensors and determining whether the display screen of the wearable device is on. Further, the method includes counting down a timeout counter when the display screen of the wearable device is on, determining whether the motion information indicates a specific pre-defined gesture when the display screen of the wearable device is off, and performing a power management function corresponding to the specific pre-defined gesture when the motion information indicates the specific pre-defined gesture.

Abstract: A face recognition method is provided. The method includes dividing an input video into different sets of frames and detecting faces of each frame in the input video. The method also includes generating face tracks for the whole video. Further, the method includes applying a robust collaborative representation-based classifier to recover a clean image from complex occlusions and corruptions for a face test sample and perform classification. In addition, the method also includes outputting the video containing the recognized face images.

Abstract: The present invention provides a salience-based media manipulation method and system for inserting contents into video presentations. The method includes establishing an advertisement-element pool containing a plurality of advertisement elements provided for insertion; obtaining an input video presenting at least one scenes and containing a plurality of original frames; and performing shot grouping and primary scene analysis on the input video to acquire scenes with static background and limited camera motion. The method also includes detecting white-space candidates in the acquired scenes for inserting at least one advertisement element and adjusting alignment between the white-space candidates and the at least one advertisement element. The adjusted alignment increases a salience gain of the advertisement elements and limit salience change of the scenes without insertion.

Abstract: A method for video program recognition includes recording video sequences of each of a plurality of channels and obtaining a set of features of an image captured by a user device. The method also includes performing a camera registration operation on the video sequences to select video frames representing the recorded video sequences based on key video frames from a registered cameras associated with the recorded video sequences. Further, the method includes extracting features of at least some of the selected plurality of video frames of each channel, storing the extracted features, and comparing the obtained set of features with the stored features of each channel to determine whether there is a match between the obtained set of features of the image and features of a video frame from a specific channel and, when there is a match, determining the specific channel as a recognized channel number.

Abstract: The present invention provides a face recognition method. The method includes obtaining a plurality of training face images which belongs to a plurality of face classes and obtaining a plurality of training dictionaries corresponding to the training face images. A face class includes one or more training face images. The training dictionaries include a plurality of deep feature matrices. The method further includes obtaining an input face image. The input face image is partitioned into a plurality of blocks, whose corresponding deep feature vectors are extracted using a deep learning network. A collaborative representation model is applied to represent the deep feature vectors with the training dictionaries and representation vectors. A summation of errors for all blocks corresponding to a face class is computed as a residual error for the face class. The input face image is classified by selecting the face class that yields a minimum residual error.

Abstract: A method is provided for content alignment. The method includes obtaining a first content sequence and a second content sequence different from the first content sequence. The method also includes representing each of the first content sequence and the second content sequence in a hierarchical structure containing an ordered root element sequence and a sub-tree structure. The ordered root element sequence includes a plurality of root elements and each root element is associated with a sub-tree of elements. The method also includes determining a desired alignment between the first content sequence and the second content sequence using dynamic programming, and outputting results of the desired alignment between the first content sequence and the second content sequence.

Abstract: A face recognition method is provided to use sparse representation and regularized least squares-based classification on a computing device. The method includes obtaining an image to be recognized as a test sample y and a set of training images of certain subjects as training sample matrix T, obtaining a sparse representation of the test sample and the training samples including an initial estimation of a sparse vector a, and constructing a new face dictionary comprising training samples with non-zero corresponding coefficients in the sparse vector a for the initial estimation. The method also includes obtaining new coefficients by solving a regularized least squares problem based on the constructed new face dictionary, and determining a face identity of the test sample based on minimum class residual calculated by using the new coefficients.

Abstract: A method for video sequential alignment is provided. The method includes inputting a first video sequence and a second video sequence, and extracting features of the first video sequence and the second video sequence. The method also includes generating an expanded path using a hierarchical sequential alignment algorithm and generating masks containing search space using keyframe matching. Further, the method includes calculating the lowest path cost within search space to find a minimal alignment path using a block A* algorithm and outputting a final alignment result after applying the block A* algorithm to search space.

Abstract: A method is provided for an intelligent video processing system based on object detection. The method includes receiving an input video sequence corresponding to a video program, obtaining a plurality of frames of the input video sequence, and obtaining a computational constraint and a temporal rate constraint. The method also includes determining one or more regions of interest (ROIs) of the plurality of frames based on the computational constraint and temporal rate constraint, and selecting a desired set of frames from the plurality of frames based on the ROIs such that the desired set of frames substantially represent a view path of the plurality of frames. Further, the method includes detecting object occurrences from the desired set of frames based on the selected desired set of frames such that a computational cost and a number of frames for detecting the object occurrences are under the computational constraint and temporal rate constraint.

Abstract: A video object retrieval method is provided. The method includes dividing an input video into a number of video shots and detecting target-class object occurrences and related-class object occurrences in each video shot. The method also includes pre-processing object bounding box images and performing automatic object extraction. Further, the method includes performing object-level feature extraction and integration and generating visual search results based on similarity scores.