Iterative Template Matching for Real-Time Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking methods, such as open source trackers and template matching, face challenges in accurately tracking non-rigid objects, handling occlusion, and adapting to object scaling and rotation, especially in real-time applications with high processing speed requirements.
Innovation Solution
An iterative template matching method that adjusts to target object appearance variations through calibration, template update, and angle detection, using a camera to generate video frames, correct perspective scaling, and refine the target object's location and orientation by narrowing down the search region and resizing the image resolution, allowing for real-time tracking without extra sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If template matching is used for tracking rigid objects, then processing speed is fast and accuracy is high, but the method fails when the template and target object differ in orientation, scale, or appearance
Solution Approach 1:
The patent applies dynamics by making the template adaptive and updateable during tracking. Instead of using a fixed template, the system dynamically updates the template based on detected target appearances, allowing the template to adapt to scale changes, rotation, and other appearance variations while maintaining fast template matching performance
Solution Approach 2:
The patent changes parameters by adjusting template resolution and search region size based on detection confidence levels. When detection is confident, the system uses higher resolution templates for accuracy; when confidence is lower, it adjusts parameters to maintain robustness, thereby adapting to different tracking conditions
2Adaptability or versatility
If open source trackers are used to track non-rigid objects, then the tracker can handle shape changes, but the processing speed is slow and accuracy to a few pixels is not achieved
Solution Approach 1:
The patent segments the tracking problem by using detection to identify candidate regions and then applying focused template matching only to those regions. This segmentation allows the system to handle complex scenarios like non-rigid objects and occlusions through detection, while maintaining fast processing by limiting expensive template matching operations to small candidate regions
3Ease of operation
If detection method is used for object location, then no previous video frame is required, but the processing speed is slow with hundreds of milliseconds latency
Solution Approach 1:
The patent applies preliminary action by performing detection on the first frame to establish initial tracking parameters and templates. Subsequent frames then use fast tracking with these pre-established parameters, achieving both the independence benefit of detection and the speed of tracking by preparing necessary information in advance
Data Source
AI summary
A method and system for tracking a position and orientation of a target object that moves along a two-dimensional plane, a camera to generate video frames of the target object, and a processor configured to correct perspective scaling of a current one of the video frames; estimate a current location and angle of the target object using a motion model or its previous location/angle; cut out a region of interest from the current video frame around the estimated current location; resize the region of interest to shrink its size by a predetermined ratio M (width=width/M, height=height/M) or equivalently reducing the image resolution of the region of interest by the ratio M (resolution=resolution/M); track a new location of the target object by template matching the current region of interest with a previously stored template; and either conclude the tracked target object location and angle are accurate and updating the template if a stopping criterion is reached with indication of successful tracking, or proceed with a refinement method to refine the tracked location and angle until the stopping criterion is eventually reached.