Dense Object Tracking via Scale-Matched Reference Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection and tracking systems in images are imprecise and computationally intensive, making them unsuitable for real-time applications like computer-generated reality (CGR) due to variations in viewing distance, angle, orientation, and lighting conditions.
Innovation Solution
A pixel-based approach for object detection and tracking that determines the scale of objects in images and uses reference images of similar scales to improve accuracy and efficiency, allowing for real-time detection and tracking by selecting or interpolating reference images based on the object's scale and adjusting the image to match the reference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing object detection and tracking techniques are used, then object detection can be performed, but the detection precision is low and computational resources are excessively consumed
Solution Approach 1:
The system performs preliminary actions by pre-processing images to generate multiple reference images at different scales before actual object detection. This preparation work includes creating a scale hierarchy from input images, which enables faster and more accurate matching during detection without consuming excessive computational resources during the actual tracking process.
Solution Approach 2:
The invention changes the parameter of image scale by generating multiple reference images at different scales and selecting the most appropriate scale for matching. This parameter transformation allows the system to adapt to objects at various distances and sizes, significantly improving detection precision while maintaining computational efficiency through scale-appropriate matching.
2Productivity
If existing object detection techniques are used, then object tracking can be performed, but the system is unsuitable for real-time applications due to computational intensity
Solution Approach 1:
The system segments the computational task by dividing objects into different scale levels and processing them separately. By creating a scale hierarchy and matching objects at appropriate scales, the system reduces overall computational complexity and enables real-time processing while maintaining lower energy consumption compared to exhaustive search methods.
Solution Approach 2:
The invention transforms the computational problem by changing the scale parameter of reference images. This allows the system to efficiently match objects at appropriate scales rather than comparing against all possible scales, dramatically reducing computational resource consumption and enabling real-time application performance.
3Measurement precision
If a single reference image is used for object detection, then the detection process is simple, but detection accuracy decreases under varying viewing conditions
Solution Approach 1:
The system implements dynamics by creating a dynamic scale hierarchy from input images and selectively using different scales based on object size and viewing distance. This dynamic adaptation allows the system to maintain high detection accuracy under varying conditions while managing reference image complexity through automated scale generation and selection rather than manual management of multiple static references.
Solution Approach 2:
The invention improves detection accuracy by changing the scale parameter of reference images to match the detected object's scale. This parameter adaptation enables accurate matching under varying viewing conditions while the system automatically manages the complexity through scale hierarchy generation and selection, rather than requiring manual management of diverse reference images.
4Measurement precision
If scale-based reference image selection is implemented, then object detection accuracy improves, but the complexity of image processing increases
Solution Approach 1:
The system performs preliminary scale hierarchy generation from input images before actual object tracking. This pre-processing creates an organized structure of reference images at different scales, which simplifies subsequent tracking operations by providing ready-to-use references at appropriate scales, thereby improving accuracy while managing processing complexity through structured preparation.
Solution Approach 2:
The invention segments the image processing task by dividing reference images into different scale levels and processing objects at their appropriate scales. This segmentation improves tracking accuracy by matching objects with references of similar scale while reducing overall processing complexity through hierarchical organization and selective processing of only relevant scale levels.
Data Source
AI summary
One exemplary implementation involves a pixel-based (also referred to as a dense-based approach) to object detection and tracking that can provide more accurate results than a feature-based approach. The efficiency of the detection and tracking is improved by using a reference image of the object that has similar characteristics (e.g., scale, lighting, blur, and the like) as the depiction of the object in the frame. In some implementations, a reference image of an appropriate scale is selected or interpolated based on the scale of the object depicted in the real world image. In other implementations, the real world image is adjusted to better match the reference image. The detection and tracking of the object can be performed with sufficient accuracy and efficiency for computer-generated reality (CGR) and other applications in which it is desirable to detect and track objects in real time.