Vehicle ROI Window Generation for Fast Attention-Based Perception
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle perception systems inefficiently allocate computing resources, leading to less accurate object detection and analysis when analyzing entire images, as they do not focus on regions of interest.
Innovation Solution
A system and method that utilize a camera device and computerized perception device to identify features in an image, assign scores based on object identification, location, and behavior, and define candidate regions of interest, focusing computing resources on these regions for enhanced analysis and path generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system analyzes entire images to ensure comprehensive object detection, then measurement precision is improved, but productivity deteriorates due to extremely low frame rates (0.2-0.5 fps)
Solution Approach 1:
The patent divides the image into multiple candidate regions of interest (ROIs) based on sensor data, rather than analyzing the entire image. This segmentation allows the system to focus computational resources on specific areas where objects are likely to be located, improving both detection accuracy and processing speed.
Solution Approach 2:
The system applies different analysis quality levels to different regions of the image. High-quality detailed analysis is applied only to selected candidate ROIs where objects are detected, while other regions receive minimal or no analysis. This local quality approach maintains detection precision for critical areas while dramatically reducing overall computational load.
2Measurement precision
If the system dedicates maximum computing resources to analyzing sensor data, then measurement precision is improved, but use of energy deteriorates due to 100% CPU usage
Solution Approach 1:
The patent segments the computational workload by identifying and analyzing only candidate regions of interest rather than processing the entire image data. This reduces CPU energy consumption from 100% to 2% while maintaining accurate object estimation in the regions that matter most for vehicle navigation.
Solution Approach 2:
The system performs partial analysis by focusing computational effort only on portions of the image where objects are detected via sensor data. Rather than exhaustively analyzing all pixels, the system applies analysis selectively to candidate ROIs, achieving sufficient precision for safety-critical applications while dramatically reducing energy consumption.
3Reliability
If the system analyzes all regions of the image to ensure complete object identification, then reliability is improved, but loss of time worsens due to insufficient frames per second
Solution Approach 1:
The patent uses sensor data to segment the image into candidate regions of interest, identifying areas where objects are likely to be located. This segmentation enables the system to maintain reliable object identification by concentrating analysis on relevant regions while processing frames at 30 fps instead of 0.2-0.5 fps.
Solution Approach 2:
The system performs preliminary filtering using sensor data to identify candidate ROIs before applying detailed image analysis. This preliminary action pre-screens the image content, ensuring that subsequent analysis is focused only on regions containing objects of interest, thereby maintaining identification reliability while reducing processing time.
Data Source
AI summary
A system for an attention-based perception includes a camera device configured to provide an image of an operating environment of a vehicle. The system further includes a computerized device monitoring the image, analyzing sensor data to identify a feature in the image as corresponding to an object in the operating environment and assign a score for the feature based upon an identification, a location, or a behavior of the object. The computerized device is further operable to define candidate regions of interest upon the image, correlate the score for the feature to the candidate regions of interest to accrue a total region score, select some of the candidate regions for analysis based upon the total region scores, and analyze the portion of the candidate regions to generate a path of travel output. The system further includes a device controlling the vehicle based upon the output.