Image Generation Apparatus for Real-Time Object Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object classification systems for autonomous driving and robotics face delays due to the need for high-resolution images and expensive equipment, such as cameras and rangefinders, which increase costs and processing time, especially when using depth information and compressive sensing techniques.
Innovation Solution
An image generation apparatus that uses light-field, compressive sensing, or coded images to identify object positions without recovering high-resolution images, allowing for real-time object classification by superimposing highlighting indications on computational images, thereby improving processing speed and reducing costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution images are used for object classification, then classification accuracy is improved, but processing time increases and costs increase
Solution Approach 1:
The patent applies partial action by performing classification on low-resolution images first to identify potential objects, then selectively recovering high-resolution images only for regions containing classified objects. This avoids the time cost of processing full high-resolution images while maintaining classification accuracy for detected objects.
Solution Approach 2:
The patent segments the image processing task into two stages: (1) classification on down-sampled low-resolution images to identify object locations, and (2) selective high-resolution recovery only for regions containing classified objects. This segmentation reduces overall processing time while preserving accuracy where needed.
2Measurement precision
If depth information is used to improve classification performance, then classification accuracy for far subjects is improved, but system cost increases due to expensive three-dimensional rangefinder
Solution Approach 1:
The patent creates a virtual depth map by processing existing two-dimensional image data through computational algorithms, copying the depth information function without requiring physical depth sensors. This provides depth-based classification capability at low cost by synthesizing depth information from standard camera images.
Solution Approach 2:
The patent replaces the mechanical three-dimensional rangefinder system with a computational approach that derives depth information from two-dimensional images using image processing algorithms. This substitution eliminates expensive hardware while achieving similar depth-based classification performance.
3Measurement precision
If compressive sensing is used to recover high-resolution images, then image resolution is improved, but calculation cost becomes enormous and real-time recovery becomes difficult
Solution Approach 1:
The patent applies partial action by performing compressive sensing recovery only on small regions containing classified objects rather than recovering entire high-resolution images. This dramatically reduces calculation cost and enables real-time processing by limiting the recovery operation to minimal necessary areas.
Solution Approach 2:
The patent segments the compressive sensing recovery operation to process only regions of interest containing classified objects, rather than recovering the entire image. This segmentation reduces computational burden from processing full high-resolution images to processing only small object regions, enabling real-time performance.
Data Source
AI summary
An image generation apparatus includes a processing circuit and a memory storing at least one computational image. The at least one computational image is a light-field image, a compressive sensing image, or a coded image. The processing circuit (a1) identifies a position of an object in the at least one computational image using a classification device, (a2) generates, using the at least one computational image, a display image in which an indication for highlighting the position of the object is superimposed, and (a3) outputs the display image.