Dynamic Vision Sensor Object Detection for Fast Motion Blur
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image sensor-based object detection methods face challenges in real-time performance due to high computational burden and motion blur when detecting fast-moving objects, especially when the object occupies a small region in the image.
Innovation Solution
A method utilizing a Dynamic Vision Sensor (DVS) to divide an image into blocks, process event data to determine moving objects, and generate candidate object rectangles, reducing computational burden and improving accuracy by identifying objects through image classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional image sensor-based object detection methods are used, then object detection can be performed, but computational burden is high and real-time performance is poor
Solution Approach 1:
The patent divides the image into multiple blocks and processes event data at the block level rather than pixel level. This segmentation approach reduces the computational burden by processing smaller units independently, enabling real-time detection while maintaining accuracy for small objects.
Solution Approach 2:
The patent extracts only the necessary event data (coordinate and timestamp) from the dynamic vision sensor and processes it directly on the image blocks where objects are located. This extraction approach eliminates unnecessary computational operations on regions without objects, significantly reducing overall computational burden.
2Reliability
If conventional object detection algorithms are used, then objects can be detected, but detection fails when objects move too fast causing motion blur
Solution Approach 1:
The patent replaces conventional image sensor-based detection with dynamic vision sensor (DVS) technology that detects object movement through changes in pixel values over time. This substitution eliminates motion blur effects because DVS captures temporal changes rather than relying on fixed-exposure images, enabling reliable detection of fast-moving objects.
Solution Approach 2:
The patent processes event data continuously over time periods, accumulating detections dynamically. This periodic processing approach allows the system to track fast-moving objects by continuously updating object positions and characteristics, maintaining detection accuracy regardless of object speed.
Data Source
AI summary
The present disclosure provides a method and a system for detecting a moving object, and a computing device. The method includes: dividing a predetermined image template into a plurality of image blocks having a same size; determining the quantity of pieces of event data in each image block within a predetermined time period, the event data being obtained by a dynamic vision sensor and triggered by movement of an object in a field of view relative to the dynamic vision sensor; determining at least one image block including the moving object in accordance with the quantity of pieces of event data in each image block; and generating a candidate object rectangle in accordance with the at least one image block, the candidate object rectangle indicating the detected moving object.