Whiteboard Atypical Event Detection Using Stroke-Based Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to effectively identify and respond to atypical events on electronic whiteboards, such as graffiti-style scribbling and drawing controversial content, due to limitations in recognizing parts of speech, interactivity, and variable sketch sequences, and lack of structural regularity in outdoor scenes and handwritten documents.
Innovation Solution
A deep learning model is employed to annotate and detect atypical events on electronic whiteboards by assigning labels to strokes, grouping contiguous sketches, and generating feature maps, with optimization techniques to enhance accuracy, and generate alerts using a rule-based method.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sketch object recognition is used to identify atypical activities, then basic object detection is achieved, but parts of speech like verbs and adjectives are not recognized
Solution Approach 1:
The patent segments the sketch recognition task into multiple components: object detection, verb recognition, and adjective recognition. Each component is handled by specific neural network modules that process different aspects of the sketch data independently, then combine results to achieve comprehensive semantic understanding.
Solution Approach 2:
The patent transitions from 2D spatial sketch data to multi-dimensional semantic representation by extracting features across multiple dimensions including object identity, verb action, and adjective properties. This dimensional expansion enables recognition of parts of speech beyond basic object categories.
2Adaptability or versatility
If stroke segmentation model is used for labelling parts in object sketches, then predefined component classes are identified, but highly variable sketch sequences are not detected
Solution Approach 1:
The patent employs dynamic neural network modules that can adapt to variable sketch sequences. The network architecture includes recurrent components and attention mechanisms that dynamically adjust to different stroke patterns and sequences, maintaining reliable detection across highly variable inputs while preserving predefined component class recognition.
3Manufacturing precision
If text localization model is used for localizing text in outdoor scenes and handwritten documents, then structured regularity is enforced, but cases without regularity are not handled
Solution Approach 1:
The patent dynamically adjusts localization parameters based on the detected structure type. For regular text regions, strict geometric constraints are applied to achieve high precision. For irregular regions, the parameters are relaxed to accommodate variable shapes and layouts, enabling the system to handle both structured and unstructured text cases effectively.
4Measurement precision
If deep learning model is trained with historical atypical events, then detection accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary actions by pre-training the deep learning model with historical atypical events offline. The model learns from historical data in advance, so during real-time operation, detection can be performed quickly without extensive processing. This separates the time-consuming training phase from the time-critical detection phase, resolving the contradiction between accuracy and processing time.
Data Source
AI summary
The embodiments herein relate to a system and a method identifying and generating an alert, using a deep learning model, for atypical events during an activity on an electronic whiteboard. The system includes a user device associated with a user, and an atypical event detecting server. The atypical event detecting server (i) receives one or more canvases at concurrent sessions, (ii) annotates the one or more canvases, (iii) detects historical atypical events from the annotated one or more canvases, (iv) trains the deep learning model by correlating the historical atypical events, historical bounding boxes with historical canvases in historical sessions, (v) processes the one or more canvases by determining a bounding box around each atypical event, and (vi) generates an alert for identified atypical events using an event record table.