Object Tracking via Camera Rotation Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object tracking technologies face challenges in accurately tracking objects in 3D space when the camera moves, as they struggle with feature extraction and matching due to camera rotation, leading to inaccurate positioning and high computational burdens, especially in real-time applications.
Innovation Solution
An electronic device and method that corrects object position using camera rotation information, applying specific correction algorithms based on noise levels and utilizing 3D projection and rotation correction to enhance tracking performance, with algorithms like Kalman filters for accurate object tracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature matching between two frames is used to track objects, then object position can be detected, but tracking accuracy deteriorates when camera rotation occurs due to blur and rolling shutter effects
Solution Approach 1:
The patent segments the object tracking problem into two independent parts: (1) camera motion estimation using IMU sensors to obtain rotation matrices, and (2) object position detection in the image plane. By separating camera motion compensation from feature matching, the system avoids the reliability issues of feature extraction under rapid rotation while maintaining tracking accuracy.
Solution Approach 2:
The patent introduces an intermediary coordinate transformation system that maps object positions from the image plane to 3D space and back, using rotation matrices as intermediaries. This intermediary transformation layer compensates for camera rotation effects, allowing accurate tracking even when direct feature matching would fail due to blur and rolling shutter effects.
2Measurement precision
If multi-camera system is used to detect object position in 3D space, then positioning accuracy improves, but system complexity and computational burden increase
Solution Approach 1:
The patent replaces the mechanical multi-camera system with a single camera combined with inertial measurement units (IMU). Instead of using multiple cameras to achieve 3D positioning, the system uses a single camera for 2D detection and compensates for 3D motion effects through mathematical transformation using rotation matrices from IMU sensors, significantly reducing system complexity.
Solution Approach 2:
The patent changes the parameter representation from direct 3D spatial coordinates (requiring multiple cameras) to a transformed coordinate system that accounts for camera rotation and translation. By parameterizing the problem in terms of rotation matrices and relative transformations, accurate 3D tracking is achieved with a single camera.
3Measurement precision
If feature extraction and matching algorithms are used for object tracking, then object position can be identified, but processing time increases making real-time tracking difficult
Solution Approach 1:
The patent performs preliminary action by pre-obtaining rotation matrices from IMU sensors before processing image frames. By having the camera motion parameters ready in advance, the system can directly transform detected 2D positions to 3D space without performing computationally intensive feature extraction and matching, enabling real-time tracking.
Solution Approach 2:
The patent extracts only the essential information needed for tracking (object center position and bounding box from simple detection) rather than performing full feature extraction. By taking out only the necessary positional data and combining it with pre-obtained rotation matrices, the system achieves accurate tracking with minimal computational burden for real-time performance.
Data Source
AI summary
An electronic device and a method for tracking an object thereof are provided. The electronic device identifies whether there is a first object being tracked, when obtaining an image and rotation information of a camera of the electronic device, corrects state information of the first object using the rotation information, when there is the first object, detects a second object matched to the first object from the image based on the corrected state information, and tracks a position of the second object using an object tracking algorithm.