• Laser & Optoelectronics Progress
  • Vol. 58, Issue 20, 2015005 (2021)
Li Ning, Jing Junfeng*, Zhang Weichuan, Bai Mengmeng, and Sun Jiurui
Author Affiliations
  • School of Electronic Information, Xi’an Polytechnic University, Xi’an, Shaanxi 710048, China
  • show less

    Abstract

    In order to solve the problem of large amount of calculation and time-consuming of the anisotropic Gaussian directional derivative filter, this paper uses the box filter to fit the anisotropic Gaussian directional derivative filter template, and proposes a new fast corner detection algorithm with excellent performance by combining template with the integral image. First, six directional derivative filter templates are designed by using box filter, and the derivative response of input image in each direction is calculated quickly by combining with integral image; second, a coarse selection mechanism of candidate pixels is proposed based on the sparsity of corners, which can quickly receive candidate pixels to reduce the number of pixels involved in subsequent operations. For each candidate pixel, the multi-directional structure tensor product is constructed by synthesizing the derivative response of each direction, and the corner measure map is generated. The performance of the proposed algorithm and 9 classical detection algorithms is evaluated under the conditions of affine transformation and Gaussian noise interference, and time-consuming comparisons are carried out on the test image set of different sizes. The experimental results show that the newly proposed algorithm has excellent detection performance and less time-consuming, and meets the needs of real-time processing.
    Copy Citation Text
    Ning Li, Junfeng Jing, Weichuan Zhang, Mengmeng Bai, Jiurui Sun. Fast Corner Detection Based on Multi-Directional Structure Tensor[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2015005
    Download Citation
    Category: Machine Vision
    Received: Nov. 2, 2020
    Accepted: Jan. 20, 2021
    Published Online: Oct. 14, 2021
    The Author Email: Jing Junfeng (jingjunfeng0718@sina.com)