Vision-Based Target Positioning Using Bundle Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the absolute position of target object points, such as in military and non-military applications, face challenges due to inaccurate metadata, reliance on expensive sensors like laser rangefinders, and the need for trained analysts and high-resolution Digital Elevation Models, making them unsuitable for time-sensitive and covert operations.
Innovation Solution
A computer-implemented method using a series of images with overlapping fields of view captured by a camera on an aircraft, analyzing image points to estimate object positions relative to the camera without requiring input orientation data, leveraging vision-based tracking and bundle adjustment techniques to derive accurate absolute positions using GPS data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If metadata from platform sensors is used to calculate target coordinates, then the calculation process is simple, but the positioning accuracy deteriorates due to inaccurate orientation data
Solution Approach 1:
The patent replaces the mechanical sensor-based orientation measurement system with a vision-based system. Instead of relying on physical orientation sensors (inertial measurement units) that provide inaccurate metadata, the system uses image analysis and feature matching to computationally determine camera orientation and target position, substituting mechanical measurement with optical measurement and mathematical processing
Solution Approach 2:
The patent introduces an intermediary computational process between image capture and target positioning. Instead of directly using sensor metadata, the system employs feature detection, feature matching across multiple images, and geometric transformations as intermediary steps to derive accurate target coordinates, mediating between the raw image data and the final positioning result
2Measurement precision
If a laser rangefinder is used to measure slant range, then accurate target position can be determined, but the system becomes expensive and inconvenient
Solution Approach 1:
The patent extracts and removes the expensive laser rangefinder component from the system. Instead of using active laser ranging, the system derives slant range information passively through geometric relationships between multiple camera positions and feature matches, extracting the necessary ranging capability from the existing camera and position data without adding expensive specialized sensors
Solution Approach 2:
The patent makes the camera system multi-functional. The same camera used for capturing images is also used for determining orientation and calculating slant range through geometric processing. Instead of requiring separate specialized sensors for each function (imaging, orientation, ranging), the system achieves multiple functions through the universal camera platform combined with computational processing
3Measurement precision
If tie points are manually defined between motion imagery and reference orthoimagery, then highly accurate coordinates can be derived, but the process requires trained analysts and is time-consuming
Solution Approach 1:
The patent implements self-service through automated feature detection and matching algorithms. Instead of requiring human analysts to manually identify and define tie points, the system automatically detects features in the motion imagery, matches them with corresponding features in reference orthoimagery, and derives coordinates without human intervention, making the process self-sufficient and eliminating the need for trained analysts
Solution Approach 2:
The patent replaces the manual human analyst process with an automated computational system. The human cognitive process of identifying and matching features is substituted with computer-based feature detection algorithms and image processing techniques, transforming a manual labor-intensive process into an automated computational process that maintains accuracy while dramatically increasing productivity
4Measurement precision
If high-resolution Digital Elevation Models are used to estimate slant range, then accurate results are achieved, but the required data may not be available
Solution Approach 1:
The patent applies partial action by using only the necessary portion of elevation information. Instead of requiring complete high-resolution Digital Elevation Models of the entire scene, the system derives sufficient elevation data from the geometric relationships in the images themselves and from available reference data, using only the minimum necessary elevation information to achieve accurate slant range estimation without requiring comprehensive external data
Data Source
Figure 1~2
Figure 3~4
Figure 5~6(d)
AI summary
The absolute position of a target object point is determined using a series of images of the scene with overlapping fields of view captured by a camera in positions arranged in at least two dimensions across the scene and position data representing the absolute positions of the camera on capture of the respective images. The images are analysed to identify sets of image points corresponding to common object points in the scene. A bundle adjustment is performed on the sets of image points that estimates parameters representing the positions of the object points relative to the positions of the camera associated with each image, but without using input orientation data representing the orientation of the camera. The absolute position of the target object point is derived on the basis of the results of the bundle adjustment and the absolute positions of the camera represented by the position data.