Object Tracking via Interest Point Clustering and Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing real-time object tracking techniques are inefficient and not robust on low-end electronic devices due to hardware constraints, leading to performance issues such as missing tracks in frame sequences and a tradeoff between tracking speed and robustness.
Innovation Solution
A method and system that extract scale and rotation invariant interest data points, cluster them, and select an optimal set based on feature response values and spread values to initiate tracking in subsequent frames, employing a clustering and grid-based mechanism for efficient and robust object tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing tracking techniques are used on low-end electronic devices, then tracking can be performed, but tracking speed and robustness deteriorate due to hardware constraints
Solution Approach 1:
The patent extracts only the most essential and informative data points from each frame for tracking purposes. Instead of processing all pixels or features, the system identifies and tracks a minimal set of key interest points that carry the most discriminative information about the object's position and appearance, thereby reducing computational load while maintaining tracking robustness
Solution Approach 2:
The patent segments the tracking problem into distinct components: interest point detection, data point selection, and tracking update. By dividing the complex tracking task into these manageable segments, the system can apply optimized algorithms to each component, improving overall efficiency and robustness on resource-constrained devices
2Loss of information
If per frame processing such as augmenting and pose estimation is performed, then more information is obtained, but real-time performance deteriorates
Solution Approach 1:
The patent applies partial action by performing only the necessary minimum processing required for effective tracking. Instead of conducting exhaustive pose estimation and augmentation on every frame, the system selectively processes only the key interest points that are most informative for tracking, achieving a balance between information extraction and processing efficiency
Solution Approach 2:
The patent performs preliminary action by pre-selecting and pre-processing the key interest points in advance before the main tracking loop. Interest points are detected and selected once, and then these pre-chosen points are tracked through subsequent frames, avoiding the need for repeated expensive detection and processing operations
Data Source
AI summary
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.