Summary

Smart Info-Media Systems in Asia

2019

Session Number:RS3

Session:

Number:RS3-4

Temporal Coherence Based Mismatch Removal for High Frame Rate and Ultra-Low Delay System

Zhe Wang,  Songlin Du,  Takeshi Ikenaga,  

pp.146-150

Publication Date:2019/9/4

Online ISSN:2188-5079

DOI:10.34385/proc.57.RS3-4

PDF download (1.8MB)

Summary:
High frame rate and ultra-low delay matching system plays an increasingly important role in human-machine interactive application. Because these applications are used in real-life complex situation which have requirement for matching accuracy, mismatch removal is needed. Although many mismatch removal algorithms have been proposed, few of them are hardware friendly and executed in real-time application with lower delay due to complicated arithmetic operations and using iterative methods. This paper proposes a hardware oriented temporal coherence based mismatch removal method for high frame rate and ultra-low delay matching system. In the proposed method, three adjacent frames are used with three times matching among each other, if the keypoint finds itself after three times matching, it is directly treated as true, which means the keypoint satisfies the temporal constraints. For the keypoints which do not find themselves, it makes statistics of matching frequency for different parts in the image to get current matching area, and uses the matching area to distinguish. Software experiment result shows that proposed mismatch removal method increases by 4.05% than basic ORB match, and by 0.33% than conventional work RANSAC. Hardware experiment result indicates that the designed image processing core successfully achieves real-time processing of 784fps VGA (640x480) video on field programmable gate array (FPGA), with a delay of 0.828 ms/frame.