Foreground Image Superimposition Angle Matching for Realistic Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in generating realistic data for object recognition models, particularly when superimposing foreground images onto background images, as they often result in unrealistic angles or postures of the target object, leading to decreased recognition accuracy.
Innovation Solution
An information processing device is developed to acquire foreground and background images, along with point cloud data and camera positions, to determine the superimposition position and calculate the background image angle, ensuring the foreground image is superimposed at an angle equal to or close to the background image angle, thereby generating realistic data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If foreground images are superimposed on background images without restrictions, then the generation process is simple and fast, but the generated data becomes unrealistic and recognition accuracy decreases
Solution Approach 1:
The patent applies parameter changes by calculating and matching the background image angle (derived from camera position and superimposition position) with the foreground image angle (from the original image metadata). By constraining the superimposition based on angle parameter matching, the system generates realistic images while maintaining automated processing efficiency.
2Adaptability or versatility
If the target object posture changes, then diverse learning data can be generated, but it becomes difficult to prepare a great amount of images of target objects in various postures
Solution Approach 1:
The patent uses copying by extracting the target object from existing images to create foreground images, then superimposing these copied objects onto different background images with matched angles. This allows generation of diverse posture images without requiring physical objects or manual image creation for each scenario.
3Reliability
If angle calculation and matching is performed to ensure realistic superimposition, then data realism improves, but the processing complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-calculating the background image angle from the camera position and superimposition position before performing the actual superimposition. The foreground image angle is extracted in advance from the original image metadata. This preliminary angle calculation and matching preparation simplifies the main superimposition process while ensuring realism.
Data Source
AI summary
An information processing device includes an acquisition unit that acquires a foreground image associated with an angle θ and indicating a target object, a background image, point cloud data, and a camera position when the background image was generated, a superimposition position determination unit that determines a superimposition position in the background image, a calculation unit that calculates a plane as a region including the superimposition position by using the point cloud data and calculates an angle θB as an angle between a straight line connecting the camera position and the superimposition position and the calculated plane, and a superimposition unit that superimposes the foreground image on the background image. The angle θ is an angle between a plane on which the target object was set and a straight line representing an image capturing direction, and is an angle equal to or close to the angle θB.