CIPV Image Resizing with Regional Downsampling for Far-Vehicle Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Advanced driver assistance systems (ADAS) face challenges in recognizing closest in-path vehicles (CIPV) due to limited computation in embedded systems, leading to reduced recognition performance and shorter detection distances for faraway vehicles, as general object recognition networks struggle with smaller object sizes and require reduced image sizes, compromising forward recognition accuracy.
Innovation Solution
An image processing system that adjusts image size by maintaining high resolution in the CIPV region and downsampling regions of lower importance, using different decimation factors for various regions based on importance, allowing more pixels for the CIPV region and fewer for others, thereby enhancing recognition performance without increasing computational costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If the image size is reduced to decrease computation, then the computational load is reduced, but the recognition performance for faraway vehicles deteriorates
Solution Approach 1:
The image is divided into multiple regions (first region, second region, third region) with different importance levels. Each region is processed with different downsampling factors, allowing the system to reduce overall computation while preserving recognition performance in critical areas like the CIPV region.
Solution Approach 2:
Different regions of the image are assigned different quality levels through varying downsampling factors. The first region (CIPV region) maintains higher quality with lower downsampling factor, while less important regions use higher downsampling factors, optimizing the balance between computation and recognition performance.
2Device complexity
If the image size is reduced to secure maximum recognition performance by increasing layers, then the number of layers can be increased, but the size of faraway vehicles is reduced lowering forward recognition accuracy
Solution Approach 1:
The image processing is segmented into region-specific downsampling operations, allowing different parts of the image to be processed at different resolutions. This enables the system to use more layers for processing the CIPV region while using fewer layers for less important regions, maintaining overall accuracy while managing computational complexity.
Solution Approach 2:
The patent applies local quality differentiation by assigning different downsampling factors to different regions. The CIPV region maintains higher quality suitable for multi-layer processing, while other regions are downsampled more aggressively, allowing the system to increase layer count where it matters most without compromising overall forward recognition accuracy.
3Productivity
If uniform downsampling is applied to reduce image size, then the computation is reduced, but the recognition rate of faraway vehicles in the own lane is lower
Solution Approach 1:
The uniform downsampling approach is replaced with segmented regional downsampling. The image is divided into multiple regions with different downsampling factors applied to each, maintaining high recognition rates in critical regions while achieving computation reduction through downsampling in less critical regions.
Solution Approach 2:
Instead of applying uniform quality reduction across the entire image, the patent implements local quality control by assigning different downsampling factors to different regions. The CIPV region maintains higher quality to preserve recognition rate, while other regions accept lower quality to improve overall computation efficiency.
Data Source
AI summary
Provided is technology to enhance a recognition distance of the own lane of a vehicle while reducing a total number of image pixels by appropriately setting a decimation factor of each region of forward image data for recognition performance enhancement of a Closest In-Path Vehicle (CIPV) region by an object recognition network.