Visual Event Detection Using Spatio-Temporal Gradients
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video processing techniques are time-consuming and prone to errors due to computational complexity and are unsuitable for detecting features and events in complex scenes or crowds with high accuracy, especially when the number of features increases.
Innovation Solution
A method involving top-down grouping and non-discrete tracking of features using spatio-temporal gradients and scene activity vectors, which locates motion edges and generates motion blur images to determine direction and velocity of objects, allowing for accurate event detection in real-time without matching features between frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If bottom-up feature detection is used to locate features in video frames, then feature detection can be performed, but the process becomes time-consuming and computationally complex
Solution Approach 1:
The patent inverts the conventional bottom-up approach by implementing a top-down feature detection method. Instead of detecting individual features and then grouping them, the system starts with predefined templates or hypotheses about objects of interest and searches for matching features in the video frames. This inversion significantly reduces computational complexity and processing time while maintaining detection accuracy.
Solution Approach 2:
The patent applies preliminary action by pre-defining templates, interest regions, or object hypotheses before video processing begins. These pre-established structures guide the feature detection process, eliminating the need for exhaustive bottom-up analysis of all pixels and regions. The preliminary setup enables faster, more targeted detection of relevant features in subsequent frames.
2Quantity of substance
If bottom-up approach detects increasing numbers of features, then more features can be tracked, but accuracy is lost when the number of features becomes large
Solution Approach 1:
The patent segments the feature detection process by dividing the video frame into predefined regions of interest or interest points based on top-down guidance. Instead of processing all pixels uniformly as in bottom-up approaches, the system focuses computational resources on specific segmented regions where features are most likely to occur. This segmentation maintains accuracy even when tracking multiple features simultaneously.
Solution Approach 2:
The patent applies local quality by assigning different detection strategies and parameters to different regions of the video frame based on their importance or likelihood of containing relevant features. High-priority regions receive more sophisticated analysis while less critical areas use simpler detection methods. This localized approach maintains overall accuracy across multiple features while optimizing processing efficiency.
3Reliability
If conventional feature matching between adjacent frames is performed, then feature tracking can be achieved, but the process is prone to errors due to lighting changes, occlusion, rotation, and scale difference
Solution Approach 1:
The patent implements dynamics by using adaptive feature detection that adjusts to changing conditions in real-time. The top-down approach dynamically updates feature templates and detection parameters based on observed scene changes, lighting variations, and object movements. This dynamic adaptation maintains reliable feature tracking across frames despite environmental changes, occlusions, rotations, and scale variations.
Solution Approach 2:
The patent applies parameter changes by modifying detection thresholds, template matching criteria, and feature selection parameters based on scene conditions. When lighting changes or occlusions occur, the system adjusts its parameters to maintain reliable feature detection and tracking. This parameter adaptation reduces errors while managing computational complexity through intelligent, condition-based adjustments.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system and method processes visual information including at least one object in motion. The visual information is processed by locating at least one spatial edge of the object, generating a plurality of spatio-temporal gradients for the at least one spatial edge over N frames, and then generating motion blur images from the spatio-temporal gradients. A regression analysis is performed on the motion blur images to determine direction of motion information of the object, and scene activity vectors are then generated for the N frames based on the direction of motion information. An event is detected in the visual information based on the scene activity vectors.