Distance-Based Image Synthesis for Hedge-Row Detection Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating training data for agricultural product detection in hedge rows is challenging due to the variety of backgrounds created by hedges, requiring significant effort and time, and resulting in erroneous detection when hedges not targeted as examination appear as background.
Innovation Solution
An information processing apparatus that acquires images and distances to detection objects, deforms background images based on object distances, and generates combined images to create training data, allowing for varied backgrounds without using hedges as object areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If training data is generated using actual captured images with hedges as background, then the training data reflects realistic conditions, but the detection model learns to detect objects from non-target hedges resulting in erroneous detection results
Solution Approach 1:
The patent creates synthetic training data by copying and combining background images with object images. Instead of using actual captured images that contain non-target hedges causing erroneous detection, the system generates artificial training images by superimposing objects onto controlled background images, thereby eliminating false detection while maintaining realistic training conditions.
2Adaptability or versatility
If a large amount of training data with various backgrounds is prepared to cover different camera angles and hedge widths, then the detection model becomes more robust, but the time and effort required to generate training data increases significantly
Solution Approach 1:
The patent performs preliminary actions by pre-processing background images and storing them for later use. Instead of manually creating diverse training data from scratch for each training scenario, the system prepares background images in advance and combines them with objects programmatically, significantly reducing the time and effort required to generate large volumes of varied training data.
Solution Approach 2:
The patent dynamically generates training data by programmatically combining objects with backgrounds using various parameters such as camera angles and hedge widths. This dynamic generation approach allows the system to create unlimited variations of training images without manual intervention, making the training data generation process adaptable and efficient.
3Productivity
If the detection model is trained to recognize detection objects in images, then the model can identify targets, but it also detects objects in non-target hedges appearing as background
Solution Approach 1:
The patent extracts the background component from actual captured images and creates separate, controlled background images. By separating the background from the object and creating synthetic combinations, the system trains the detection model to focus only on target objects against controlled backgrounds, preventing the model from learning to detect objects in non-target hedges while maintaining its ability to detect targets.
Data Source
AI summary
An information processing apparatus includes a first acquisition unit configured to acquire a first image in which a detection object is captured, and a distance to the detection object in the first image, a second acquisition unit configured to acquire a second image, a first generation unit configured to deform the second image based on the distance to the detection object in the first image, and generate a combined image based on the first image and the deformed second image, and a second generation unit configured to generate training data by using the combined image.