Region-Selective Image Pre-Processing for Autonomous Driving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image compression techniques lack the ability to control compression based on individual regions of an image, leading to suboptimal compression and interference with image analysis algorithms, particularly in applications like remote and autonomous driving where certain regions are more critical than others.
Innovation Solution
A method that filters images to reduce information content in non-critical regions before encoding, using techniques such as object recognition, distance values, temperature values, or importance matrices to determine regions of interest and apply appropriate filtering, thereby reducing computational costs and file size.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If overall compression is applied to the entire image, then the data transmission bandwidth is reduced, but the image quality in critical regions deteriorates and compression artifacts interfere with image analysis algorithms
Solution Approach 1:
The patent applies different compression qualities to different regions of the image based on their importance. Critical regions (containing objects of interest for autonomous driving) are compressed with lower intensity to preserve quality, while non-critical regions are compressed more aggressively. This is achieved by generating an importance map that identifies critical regions and applying region-specific compression parameters, thereby resolving the contradiction between overall bandwidth reduction and preservation of critical image quality.
Solution Approach 2:
The patent segments the image into multiple regions based on importance - critical regions containing objects of interest and non-critical regions. By dividing the image processing into separate regions with different compression treatments, the system can optimize bandwidth usage while preserving quality where needed, directly addressing the technical contradiction.
2Device complexity
If uniform compression is applied to all regions, then the encoding process is simple, but the compression efficiency is suboptimal and does not prioritize critical information
Solution Approach 1:
The patent performs preliminary action by generating an importance map before compression to identify critical regions. This pre-processing step enables the subsequent compression process to efficiently target critical areas, improving compression efficiency without requiring complex real-time adjustments during encoding. The importance map generation prepares the data structure needed for optimized compression.
3Speed
If high compression is applied to reduce file size, then transmission speed increases, but loss of information increases and image analysis accuracy deteriorates
Solution Approach 1:
The patent applies local quality differentiation by preserving high information quality in critical regions while allowing higher compression in non-critical regions. This enables faster overall transmission compared to uniform high compression, while maintaining sufficient image information quality in regions that matter for autonomous driving decisions, thus resolving the contradiction between transmission speed and information preservation.
Data Source
Figure 1~2
Figure 3A~3B
Figure 4A~4B
AI summary
The present invention relates to systems and methods for pre-processing images. In particular, image processing that is performed on images that are recorded by a camera of a vehicle. A method, system and computer-readable medium described herein provide one or more images with reduced information content. In particular, the one or more images are filtered to generate a filtered image with reduced information content before said filtered image is encoded. This may lead to a decrease in computational steps performed by an encoder when encoding the filtered image as well as to a decrease in the file size of the encoded image that needs to be stored and/or transmitted. One or more images with reduced information content are provided before encoding the filtered image by receiving an image having a plurality of pixels, determining a region of interest within the image and filtering the image to generate a filtered image with reduced information content. The filtering comprises reducing the information content of the image by filtering pixels of the image outside the region of interest. The method further comprises encoding the filtered image.