A fast corner detection method for time domain vision sensor signal processing

Abstract

The invention discloses a fast corner detection method for signal processing of a time domain vision sensor, Different from 'frame sampling' imaging method, vision sensor is sensitive to light intensity change in shooting scene, and its output visual information has the characteristics of real-time, very low redundancy and small amount of data. Compared with the traditional frame sampling imagingmethod, the processing and storage requirements are greatly reduced, so it is more suitable for high-speed vision applications. The invention adopts an event-driven mode, and judges whether each inputevent is generated by a corner point according to the temporal-spatial correlation between the events. In the calculation process, the 'local closest event distribution map' centered on the input event is first constructed, and then the 'local closest event distribution map' is partitioned and statistically and binarized. According to the obtained partition binary representation, the table-lookupmethod is used to determine whether the input event is generated by the corner point or not. In addition to corner detection, this method can simultaneously judge the direction and type of corner points, and provide more abundant feature information for further target recognition.

Description

technical field [0001] The invention belongs to the technical field of machine vision, and in particular relates to a fast corner detection method for time domain vision sensor signal processing. Background technique [0002] Among all human information sources, image information accounts for more than 80%. At present, silicon-based semiconductor image sensors (including CCD and CMOS image sensors) have completely replaced silver iodide film and become the most important photoelectric imaging device. Based on the digital images it generates, computer vision and machine vision have been used more and more widely, and have become an important part of artificial intelligence. [0003] According to the imaging principle, the currently used images are generated based on the "frame sampling" method: [0004] 1. After all pixels are reset at the same time, they start to receive light (collect photoelectric charges), and stop receiving light after reaching the set exposure time; ...

AI Technical Summary

Problems solved by technology

These redundant information also put huge pressure on the processing and storage of the system;

[0010] (2) High latency

Due to the need for gradient calculation, the amount of calculation is large and the real-time performance is poor; an event-driven focus detection method is designed based on the principle of the FAST operator, and corner detection is performed on two concentric circles

Although this method avoids the gradient calculation, it has a large amount of calculation and weak anti-noise performance due to a large number of continuous similar pixel searches.

Images