UAV Visual Tracking With Local Re-Identification and FFT Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing consumer UAVs face challenges in implementing high-performance visual tracking without GPS, as current solutions require powerful computational platforms, making them unsuitable for low-end consumer UAVs.
Innovation Solution
A real-time visual object tracking system that uses correlation maps computed between image patches within search windows, employing feature extraction and Fast Fourier Transform to determine the updated location of a target object, allowing for low-complexity and high-performance tracking on low-end platforms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If high-complexity visual tracking functionalities are implemented, then tracking performance is improved, but device complexity increases requiring powerful computational platforms
Solution Approach 1:
The patent segments the visual tracking task into distinct modules: feature extraction, correlation map computation, and target location determination. This modular approach allows each component to be optimized independently and executed efficiently on resource-constrained embedded platforms, resolving the contradiction between tracking performance and device complexity
Solution Approach 2:
The patent uses template matching where a reference image patch (template) is copied and compared against regions in the target video frame. This copying approach enables robust visual tracking without requiring complex machine learning models, achieving reliable tracking performance on low-end devices
2Adaptability or versatility
If real-time visual tracking is implemented without GPS, then tracking capability is improved for GPS-denied environments, but computational requirements increase
Solution Approach 1:
The patent replaces GPS-based mechanical positioning with vision-based optical tracking. By substituting the GPS mechanical system with a camera and image processing algorithm, the system achieves adaptability in GPS-denied environments while maintaining reasonable energy consumption through efficient correlation computations
3Measurement precision
If feature extraction and correlation map computation are used, then tracking accuracy is improved, but processing time increases
Solution Approach 1:
The patent implements tracking at periodic intervals rather than continuously processing every video frame. By selecting key frames for feature extraction and correlation computation, the system maintains high tracking accuracy while significantly reducing overall processing time and energy consumption
Solution Approach 2:
The patent applies correlation map computation only to regions of interest within video frames rather than processing the entire frame. This partial action approach maintains tracking accuracy in the target region while reducing computational load and processing time
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables effective visual tracking of moving objects without GPS, achieving real-time performance on low-end UAVs by simplifying the tracking process through feature-based correlation methods, thus expanding the applicability of visual tracking capabilities.
Implementation Method 1
employing feature extraction and Fast Fourier Transform to determine the updated location of a target object
Data Source
AI summary
Embodiments described herein provide various examples of real-time visual object tracking. In another aspect, a process for performing a local re-identification of a target object which was earlier detected in a video but later lost when tracking the target object is disclosed. This process begins by receiving a current video frame of the video and a predicted location of the target object. The process then places a current search window in the current video frame centered on or in the vicinity of the predicted location of the target object. Next, the process extracts a feature map from an image patch within the current search window. The process further retrieves a set of stored feature maps computed at a set of previously-determined locations of the target object from a set of previously-processed video frames in the video. The process next computes a set of correlation maps between the feature map and each of the set of stored feature maps. The process then attempts to re-identify the target object locally in the current video frame based on the set of computed correlation maps.


