China
Africa
Explore chapters and articles related to this topic
Image Descriptors and Features
Published in Manas Kamal Bhuyan, Computer Vision and Image Processing, 2019
Harris corner detector Harris corner detector is a corner detection operator which is commonly used to extract corners of objects present in an image. This corner detector is more accurate in distinguishing between edges and corners. The basic principle of this detector is the detection of image intensity changes around a point. For this, the auto-correlation matrix is used. The auto-correlation matrix employs first-order image derivatives. Originally, the image derivatives were computed by small filters. However, Gaussian filters are more suitable for this operation as Gaussian function can do additional image smoothing to suppress noises. The type of image pattern present inside the window around a given image point can be decided on the basis of eigenvalues of the auto-correlation matrix M. Two large eigenvalues determine the presence of a corner point. Also, one large eigenvalue indicates an edge pixel. Hence, the Harris corner detector employs a corner response function, which is represented in terms of the auto-correlation matrix.
A marker-free method for structural dynamic displacement measurement based on optical flow
Published in Structure and Infrastructure Engineering, 2021
Jinsong Zhu, Ziyue Lu, Chi Zhang
In the field of civil engineering, however, the precision of KLT often depends on the quality of key points. For choosing key points with an appropriate amount and robust properties, some corner detectors are used. The most popular methods are Harris corner detector (Harris et al., 1988), Shi–Tomasi corner, scale-invariant feature transform (SIFT) (Lowe, 2004) and speed-up robust features (SURF) (Dong, Celik, & Catbas, 2019). However, one of the main drawbacks of optical flow is the drift (Brox & Malik, 2011; Li, Cosker, & Brown, 2012). Errors accumulate between frames over time result in the movement away from the correct tracking trajectory. This problem may be not noticeable between single image pairs, but accumulation can be particularly problematic when tracking across long sequences (Li, Ren, et al., 2016). There are still challenges in choosing key points with the balance of quality and amount.
Three-dimensional CityGML building models in mobile augmented reality: a smartphone-based pose tracking system
Published in International Journal of Digital Earth, 2021
Christoph Blut, Jörg Blankenbach
As Table 1 shows, the Harris corner detector in average finds the most true corners, but on the downside also detects the most false corners, which are ∼20 times higher than with the Shi-Tomasi detector or curvature-based detector. This is crucial for the detection speed of the door detector, since larger corner numbers, generally, result in longer detection runtimes. Comparing the Shi-Tomasi detector to the curvature-based detector, the results are quite similar, with equal average found true corners and only slight differences in the amount of detected false corners. The Shi-Tomasi detector in average detects less false corners, which speaks for this solution, but the curvature-based detector showed better performance for the specific case of door detection, in which for 89% of the evaluation images the door corners were included. Therefore, the curvature-based detector was used for our implementation of the door detection algorithm.
A triangle mesh-based corner detection algorithm for catadioptric images
Published in The Imaging Science Journal, 2018
In computer vision, corners are image points which has a well-defined position and can be robustly detected. Corners are widely used in many computer vision problems, such as image matching, object tracking, 3D reconstruction, and SLAM. A large number of corner detection algorithms have been defined based on various extraction techniques. Among proposed methods Harris corner detector [6] is a well-known algorithm which is based on image derivatives. The Harris detector uses the second moment matrix, also called the autocorrelation matrix, to describe the variation of light intensity in a local neighbourhood of an image point:where and are differentiation and integration scales, respectively; andThe eigenvalues of the autocorrelation matrix correspond to the principal light intensity changes in two orthogonal directions in a neighbourhood of an image point. Therefore, corners can be detected as image points for which the variations of intensity are significant in both directions.