Object Tracking in Incomplete Disparity Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object tracking methods fail when the size of the tracked object is unknown and the disparity map is incomplete, leading to incorrect tracking results due to incomplete disparity maps and incorrect detection of object size and position.
Innovation Solution
An object tracking method that predicts the position of the object based on historical frames, corrects the detected position using a confidence level, and adjusts the center of the object by deriving candidate centers to improve accuracy in incomplete disparity maps.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If template matching tracking is used to track objects at known positions and depths, then tracking accuracy is improved for known objects, but the method cannot be applied when object size and template positions are unknown
Solution Approach 1:
The patent changes the parameter representation from fixed templates to dynamic parameter sets. Instead of using predetermined templates for known positions and depths, the system represents objects with parameters including position, depth, size, and shape that can adapt to unknown objects. This allows the tracking system to handle variable object characteristics rather than requiring pre-defined templates.
Solution Approach 2:
The patent creates a universal tracking method that works for both known and unknown objects. By using parameter-based representation instead of template matching, the system achieves multi-functionality - it can track objects with known characteristics as well as discover and track objects with unknown characteristics, making the system versatile across different scenarios.
2Device complexity
If prediction-detection-correction method based on center is used, then tracking is simplified, but incorrect tracking results occur when disparity map is incomplete
Solution Approach 1:
The patent adds depth as another dimension to the tracking parameters. Instead of only using 2D center-based prediction, the system incorporates depth information from the disparity map to predict object position in 3D space. This dimensional enhancement allows the system to maintain reliability even when parts of the disparity map are incomplete, as depth provides additional spatial context.
Solution Approach 2:
The patent implements a correction mechanism that uses feedback from the disparity map to adjust predicted positions. The system detects objects in the current frame, compares detected positions with predicted positions, and corrects deviations. This feedback loop maintains tracking reliability by continuously refining position estimates based on actual detection results, compensating for incomplete disparity map data.
3Productivity
If only detected parts of object are used for tracking, then processing is simplified, but object position and size are incorrectly determined
Solution Approach 1:
The patent performs preliminary prediction of object position and size based on previous frames before detecting the current frame. By predicting where the object should be and what its characteristics should be, the system can then use this prior information to guide the detection process and interpret partial disparity map data more accurately, maintaining measurement precision without excessive processing complexity.
Solution Approach 2:
The system uses feedback from predicted object parameters to guide the detection and correction process. When only parts of the object are detected due to incomplete disparity maps, the predicted parameters provide reference values that help correct the detected partial information, ensuring accurate determination of complete object position and size.
Data Source
Figure 1~2
Figure 3(a)~3(c)
Figure 4~5
AI summary
A method and an apparatus for tracking an object are disclosed. The method includes the steps of detecting, from a disparity map of a current frame, a position of the object including a size of the object and a center of the object; predicting the position of the object in the disparity map of the current frame based on the positions of the object detected from the disparity maps of a predetermined number of frames immediately before the current frame; and correcting the position of the object detected from the disparity map of the current frame to obtain a corrected position based on a confidence level of the size of the object detected from the disparity map of the previous frame, if the position of the object detected from the disparity map of the current frame matches the predicted position of the object.