Machine Vision Compression ROI Quantization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image and video compression technologies, optimized for human perception, fail to efficiently manage the varying importance of objects in video frames, leading to suboptimal compression efficiency in machine perception tasks such as autonomous driving and remote driving, where not all objects require equal quality and compression.
Innovation Solution
A computer-implemented method and system that detects regions of interest (ROI) in video frames, determines a partitioning scheme and quantization parameter based on machine learning algorithms, and selects a quantization parameter table tailored to specific machine vision tasks using cost function optimization to improve coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If uniform compression is applied to all regions of video frames, then implementation simplicity is maintained, but compression efficiency for machine perception deteriorates because not all objects require equal quality
Solution Approach 1:
The video frame is segmented into multiple regions based on object importance. The system identifies key objects (e.g., pedestrians, vehicles, traffic signs) and divides the frame into regions of interest (ROI) and non-ROI areas. Different compression algorithms and parameters are applied to each segment, with higher quality preservation for ROI regions containing critical objects and more aggressive compression for non-critical areas, thereby improving overall compression efficiency for machine perception tasks.
Solution Approach 2:
The patent implements local quality variation by applying different quantization parameters, transformation types, and compression strengths to different spatial regions within the video frame. Regions containing important objects for machine perception (such as road users, obstacles, and traffic signals) receive higher quality encoding with finer detail preservation, while background or less critical regions undergo more aggressive compression, optimizing the trade-off between compression ratio and perception task performance.
2Productivity
If higher compression rates are applied, then data transmission and storage efficiency improves, but machine perception performance deteriorates due to loss of critical visual information
Solution Approach 1:
The system incorporates feedback mechanisms where the compression process is guided by the specific machine perception tasks to be performed (e.g., object detection, semantic segmentation, tracking). The algorithm analyzes the content of video frames, identifies objects relevant to the perception tasks, and adjusts compression parameters dynamically to preserve critical visual features needed for accurate perception while achieving high compression rates for non-critical regions.
Solution Approach 2:
The patent dynamically changes compression parameters based on scene content and task requirements. Different quantization parameters (QP), transformation block sizes, and prediction modes are selected for different regions and frames. For example, frames containing critical objects or high-motion areas use lower compression parameters, while static or less important frames use higher compression, thereby maintaining perception accuracy while maximizing overall compression efficiency.
3Productivity
If region-based differential compression is implemented, then compression efficiency for machine perception improves, but computational complexity increases due to ROI detection and parameter optimization
Solution Approach 1:
The system performs preliminary analysis of video frames to identify regions of interest and determine appropriate compression parameters before the actual compression process. Object detection, classification, and ROI segmentation are executed in advance, allowing the compression algorithm to be configured optimally for each region. This preliminary action enables the system to prepare compression strategies based on content analysis, improving coding efficiency while managing computational complexity through structured preprocessing.
Data Source
AI summary
A method and a system described herein provide optimizing image and/or video compression for machine perception. According to an aspect, the method comprises receiving a raw image frame from a camera sensor; detecting a predefined object in the raw image frame and marking a region around the predefined object within the raw image frame as ROI. Based on the ROI, a partitioning scheme, a prediction mode, and quantization parameter are determined for improving coding efficiency. Machine perception efficiency is improved by selecting a quantization parameter table used for compressing and encoding the raw image or video frame based on a selected machine vision task. The selection of the quantization parameter table is based on training of the selected machine vision task using cost function optimization.