Hierarchical Image Capture for Low-Latency Object Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer vision tasks in image processing pipelines face challenges due to high power consumption and latency issues when using always-on image sensors and intelligent controllers, particularly in power-constrained systems, where scene content can change between initial identification and subtasks, leading to inefficiencies in capturing high-resolution images.
Innovation Solution
A method for low latency hierarchical image capture involves capturing a first image, detecting regions-of-interest, determining image characteristics, and capturing a set of second images based on these characteristics, which are then combined to recognize objects, optimizing power usage and reducing latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple high-resolution images are captured to improve computer vision task performance, then recognition accuracy is improved, but power consumption increases and latency increases
Solution Approach 1:
The image capture process is segmented into two stages: first capturing a low-resolution overview image to identify regions-of-interest, then capturing only high-resolution images for those specific regions. This segmentation allows the system to achieve high recognition accuracy for important objects while minimizing overall power consumption by avoiding full high-resolution capture of the entire scene.
Solution Approach 2:
The system applies different quality levels to different regions of the image. Regions-of-interest are captured at high resolution while other areas are captured at low resolution or skipped entirely. This local quality approach maintains recognition accuracy for critical objects while reducing the total power consumption compared to uniform high-resolution capture.
2Measurement precision
If multiple high-resolution images are captured to improve computer vision task performance, then recognition accuracy is improved, but processing latency increases
Solution Approach 1:
The capture sequence is segmented so that low-resolution overview images are captured first to quickly identify regions-of-interest, followed by targeted high-resolution captures. This segmentation reduces total latency because the system avoids capturing unnecessary high-resolution images for the entire scene, focusing computational resources only on relevant regions.
Solution Approach 2:
The system performs preliminary low-resolution image capture and region-of-interest detection before committing to high-resolution capture. This preliminary action allows the system to pre-identify which regions require high-resolution processing, thereby reducing overall processing latency by avoiding unnecessary high-resolution captures.
3Ease of operation
If always-on image sensors and intelligent controllers are used to facilitate image capture, then image capture capability is improved, but power consumption increases
Solution Approach 1:
Instead of using always-on image sensors, the system employs periodic image capture triggered by specific events or conditions. Low-resolution images are captured periodically to monitor the scene, and high-resolution capture is triggered only when regions-of-interest are detected, thereby maintaining operational capability while significantly reducing power consumption compared to continuous high-resolution capture.
Data Source
AI summary
Features described herein pertain to low latency hierarchical image capture. An image can be captured using a pixel array of an image sensing system. A region-of-interest (ROI) can be detected in the image. Image characteristics of the image can be determined based on the region-of-interest (ROI). Determining the image characteristics of the image can include determining an image quality level of the region-of-interest (ROI). Image capturing instructions for capturing a set of images can be determined based on the image characteristics. The set of images can be combined into an image and an object can be recognized in the image.