Event-Based Vision Sensor Image Alignment and Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Event-based vision sensors generate images with lower quality compared to frame-based sensors due to their response to light changes, resulting in blurry images, especially under low illumination and with hand shake, and the quality of generated images is compromised when collecting events over time.
Innovation Solution
An image generating method that aligns first and second images from event-based vision sensors using motion, rotation, and enlargement, determining thresholds based on dispersion and solid levels of event signals, and extracting additional signals to improve image quality, while removing noise and correcting lens distortion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If images are generated by collecting events from event-based vision sensor over time periods, then processing speed increases and data amount decreases, but image quality becomes lower compared to frame-based vision sensor
Solution Approach 1:
The patent divides the image generation process into multiple time periods (first time period, second time period, etc.) and generates separate images for each period. Events are segmented by time period and processed independently to create multiple images that are later aligned and combined. This segmentation allows the system to maintain high processing speed while improving image quality through temporal diversity.
Solution Approach 2:
The patent merges multiple images generated from different time periods by aligning them based on motion information. The aligned images are then combined to produce a final high-quality image. This merging process integrates information from multiple time periods, compensating for the limitations of single-period event-based imaging while maintaining the speed advantages of event processing.
2Productivity
If event-based vision sensor responds to light changes to detect contours, then high-speed processing is achieved with small data amount, but only contour information is captured without full image quality
Solution Approach 1:
The patent performs preliminary actions by generating multiple images from different time periods before final alignment and combination. By pre-processing events into multiple temporal segments and creating intermediate images, the system prepares diverse information that will be integrated later, preserving detail information that would be lost in single-period processing while maintaining event-based speed advantages.
Solution Approach 2:
The patent maintains continuous useful action by continuously collecting and processing events across multiple time periods without interruption. The overlapping time periods ensure continuous coverage and information accumulation, allowing the system to maintain high processing speed while progressively building complete image information through continuous event stream processing.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image generating method and an image generating apparatus are provided. The image generating method includes generating a first image based on first event signals that are output from an event-based vision sensor during a first time period and stored in a buffer, generating a second image based on second event signals that are output from the event-based vision sensor during a second time period and stored in the buffer, and aligning the first image and the second image to generate a third image having a quality that is higher than qualities of the first image and the second image.