Abstract: The present invention discloses method and virtual assistance system for determining response to user queries. The virtual assistance system receives data comprising plurality of interaction between one or more users and one or more real agents for resolving query, where data is classified into user and real agent data. Entities and intent identified from each of the classified user data and real agent data are classified using predefined domain model. The entities and intent are combined to identify plurality of sequence of resolution data. Based on classification of entities and intent, virtual assistance system determines first set of resolution data and second set of resolution data. Thereafter, each of plurality of sequence of resolution data is clustered into one or more category associated with type of users, based on first set and second set of resolution data, parameters associated with users and historical resolution data used for responding to query.

Abstract: This disclosure relates to method and system for of dynamically annotating data or validating annotated data. The method may include receiving input data comprising a plurality of input data points. The method may further include one of: a) generating a plurality of annotations for each of the plurality of input data points using at least one of a state-label mapping model and a comparative ANN model, or b) receiving the plurality of annotations for each of the plurality of input data points from an external device or from a user, and validating the plurality of annotations using at least one of the state-label mapping model and the comparative artificial neural network (ANN) model.

Abstract: Methods and content rendering systems for rendering Augmented Reality (AR) content for textureless objects are disclosed. The content rendering system captures one or more image frames associated with objects and transmits to a cloud server to detect one or more textureless objects along with one or more associated buffer Regions of Interest (ROI) based on a pre-trained object detection model. A grid index associated with each buffer ROI defines a position of the one or more textureless objects in each buffer ROI. The content rendering system obtains AR script from the cloud server based on the detected one or more textureless objects. Thereafter, based on the AR script, AR content associated with the one or more textureless objects is rendered in real time at the position of the one or more textureless objects.

Abstract: A method and a system are described for multi-modal input based platform for intent based product recommendations. The method comprises receiving, by the product recommendation device, one or more multi-modal user inputs associated with a first product, wherein the one or more multi-modal user inputs are at least speech, text, bodily expressions and clickstream data. It further includes determining an intent-score for each of the one or more multi-modal user inputs using one or more trained data models. It further includes computing an emotion-score by aggregating the intent-score for each of the one or more multi-modal user inputs based on a weighted average of the intent-score for each of the one or more multi-modal user inputs and the method then includes recommending one or more second products based on the emotion-score.

Abstract: The present invention relates to a method of providing context based personalized assistance. The method includes receiving a request for assisting a user for performing a task. A workflow to the user is provided based on the received request. A multimodal data associated with one or more actions performed by the user while performing each step of the workflow is captured and a context of the one or more actions and an instance of struggle by the user in performing the one or more actions is determined using a machine learning technique. A personalized recommendation for the user for performing the one or more actions is identified in real time and the personalized recommendation is provided to the user for completing the task.

Abstract: The present invention discloses method and virtual assistance system for determining response to user queries. The virtual assistance system receives data comprising plurality of interaction between one or more users and one or more real agents for resolving query, where data is classified into user and real agent data. Entities and intent identified from each of the classified user data and real agent data are classified using predefined domain model. The entities and intent are combined to identify plurality of sequence of resolution data. Based on classification of entities and intent, virtual assistance system determines first set of resolution data and second set of resolution data. Thereafter, each of plurality of sequence of resolution data is clustered into one or more category associated with type of users, based on first set and second set of resolution data, parameters associated with users and historical resolution data used for responding to query.

Abstract: This disclosure relates to method and system for of dynamically annotating data or validating annotated data. The method may include receiving input data comprising a plurality of input data points. The method may further include one of: a) generating a plurality of annotations for each of the plurality of input data points using at least one of a state-label mapping model and a comparative ANN model, or b) receiving the plurality of annotations for each of the plurality of input data points from an external device or from a user, and validating the plurality of annotations using at least one of the state-label mapping model and the comparative artificial neural network (ANN) model.

Abstract: This disclosure relates to method and system for detecting and correcting an orientation of an image. The method may include generating a number of images of an environment corresponding to the input image, determining a region and a label corresponding to each of a number of objects in each of the images, determining a context for each of the objects in each of the images with respect to an overall context of the environment based on the label for each of the objects, estimating an orientation score for each of the images based on a correlation between the region and the label of each of the objects in each of the images and the context of each of the objects in each of the images, and correcting the orientation of the input image based on the estimated orientation score for each of the images.

Abstract: Methods and content rendering systems for rendering Augmented Reality (AR) content for textureless objects are disclosed. The content rendering system captures one or more image frames associated with objects and transmits to a cloud server to detect one or more textureless objects along with one or more associated buffer Regions of Interest (ROI) based on a pre-trained object detection model. A grid index associated with each buffer ROI defines a position of the one or more textureless objects in each buffer ROI. The content rendering system obtains AR script from the cloud server based on the detected one or more textureless objects. Thereafter, based on the AR script, AR content associated with the one or more textureless objects is rendered in real time at the position of the one or more textureless objects.

Abstract: A method and device for detecting region of interest based on images is disclosed. The method includes creating, by a region detection device, at least one histogram associated with at least one storage area within an image captured for a predetermined location. The method further includes identifying, by the region detection device, a plurality of boundaries within the at least one storage area based on the at least one histogram. The method includes detecting, by the region detecting device, at least one region of interest based on the plurality of boundaries.

Abstract: A method and a system are described for multi-modal input based platform for intent based product recommendations. The method comprises receiving, by the product recommendation device, one or more multi-modal user inputs associated with a first product, wherein the one or more multi-modal user inputs are at least speech, text, bodily expressions and clickstream data. It further includes determining an intent-score for each of the one or more multi-modal user inputs using one or more trained data models. It further includes computing an emotion-score by aggregating the intent-score for each of the one or more multi-modal user inputs based on a weighted average of the intent-score for each of the one or more multi-modal user inputs and the method then includes recommending one or more second products based on the emotion-score.

Abstract: Disclosed herein is a method and system for automating context-based switching between user activities. The method comprises receiving user inputs from users in one or more input modes and based on a predefined score associated with input modes the context of user input is determined. Based on the determined context and one or more predefined activities of users, a predefined first activity is recommended. Further, a deviation from the context associated with the predefined first activity to a context associated with a second activity is detected, based on user input. Since the context is deviated, the second activity is performed upon detecting availability of pre-recorded activities related to second activity. Once the second activity is completed, system switches back to predefined first activity to complete predefined first activity. The system understands context of the user based on user inputs and performs different activities according to users comfort and requirements.

Abstract: This disclosure relates generally to computer vision, and more particularly to method and system for tracking objects within a video. In one embodiment, a method for tracking objects within a video is disclosed. The method includes receiving one or more regions of interest (ROIs) corresponding to one or more objects in an initial frame of the video, extracting a set of scale and rotation invariant interest data points in each of the ROIs, clustering the set of scale and rotation invariant interest data points in a ROI into a set of clusters based on corresponding locations in the ROI, determining an optimal set of interest data points from each of the set of clusters based on corresponding feature response values and spread values, and initiating tracking of the optimal set of interest data points in subsequent frames of the video to track the one or more objects in the video.

Abstract: Disclosed herein is a method and system for identifying type of an input document in real-time. In an embodiment, visual features and keywords of the input document are compared with reference visual features and reference keywords extracted from plurality of predetermined document types for computing a relative similarity score for the input document. Subsequently, one or more best-match document types are identified among the plurality of predetermined document types based on the relative similarity score of the input document. Thereafter, visual features and keywords of the input document are compared with global and local characteristics of the best-match document types for identifying the type of the input document. In an embodiment, the present disclosure helps in recognizing type of a document prior to digitizing the document, and thereby helps in storing the digitized documents in correct formats and appropriate storage directories.

Abstract: This disclosure relates to method and system for detecting and correcting an orientation of an image. The method may include generating a number of images of an environment corresponding to the input image, determining a region and a label corresponding to each of a number of objects in each of the images, determining a context for each of the objects in each of the images with respect to an overall context of the environment based on the label for each of the objects, estimating an orientation score for each of the images based on a correlation between the region and the label of each of the objects in each of the images and the context of each of the objects in each of the images, and correcting the orientation of the input image based on the estimated orientation score for each of the images.

Abstract: Disclosed subject matter relates to digital media including a method and system for generating a contextual audio related to an image. An audio generating system may determine scene-theme and viewer theme of scene in the image. Further, audio files matching scene-objects and the contextual data may be retrieved in real-time and relevant audio files from audio files may be identified based on relationship between scene-theme, scene-objects, viewer theme, contextual data and metadata of audio files. A contribution weightage may be assigned to the relevant and substitute audio files based on contextual data and may be correlated based on contribution weightage, thereby generating the contextual audio related to the image. The present disclosure provides a feature wherein the contextual audio generated for an image may provide a holistic audio effect in accordance with context of the image, thus recreating the audio that might have been present when the image was captured.

Abstract: This disclosure relates generally to the text-to-speech synthesis and more particularly to a system and method for rendering textual messages using customized natural voice. In one embodiment, a system for rendering textual messages using customized natural voice, is disclosed, comprising a processor and a memory communicatively coupled to the processor. The memory stores processor instructions, which, on execution, causes the processor to receive present textual messages and at least one of previous textual messages, response to the previous textual messages or receiver's context. The processor further predicts final emotional state of sender's customized natural voice based on an intermediate emotional state and the receiver's context. The processor further synthesizes the sender's customized natural voice based on the predicted final emotional state of the sender's customized natural voice, voice samples and voice parameters associated with the at least one sender.

Abstract: The present disclosure relates to method and system for determining intent of subject using behavioural pattern by intent determination system. The intent determination system receiving video data associated with subject from one or more information sources, extracting pre-defined number of femes before target frame from video data of subject, obtaining plurality of behavioural parameters of subject for each of extracted frames, determining score for plurality of behavioural parameters of extracted femes using trained intent model, calculating weighted average score for plurality of behavioural parameters of extracted frames based on score of each of behavioural parameters, analysing emotion of subject for pre-defined number of frames based on weighted average score and determining intent of subject based on emotion of subject analysed. The present disclosure removes the need of human intervention in determining intent of subject.

Abstract: This disclosure relates generally to computer vision, and more particularly to method and system for tracking objects within a video. In one embodiment, a method for tracking objects within a video is disclosed. The method includes receiving one or more regions of interest (ROIs) corresponding to one or more objects in an initial frame of the video, extracting a set of scale and rotation invariant interest data points in each of the ROIs, clustering the set of scale and rotation invariant interest data points in a ROI into a set of clusters based on corresponding locations in the ROI, determining an optimal set of interest data points from each of the set of clusters based on corresponding feature response values and spread values, and initiating tracking of the optimal set of interest data points in subsequent frames of the video to track the one or more objects in the video.

Abstract: A method and device for detecting region of interest based on images is disclosed. The method includes creating, by a region detection device, at least one histogram associated with at least one storage area within an image captured for a predetermined location. The method further includes identifying, by the region detection device, a plurality of boundaries within the at least one storage area based on the at least one histogram. The method includes detecting, by the region detecting device, at least one region of interest based on the plurality of boundaries.