Fast Library For Approximate Nearest Neighbors Research Articles

A Large-Crack Image-Stitching Method with Cracks as the Regions of Interest

While crack detection is crucial for maintaining concrete structures, existing methods often overlook the analysis of large cracks that span multiple images. Such analyses typically rely on image stitching to create a complete image of a crack. Current stitching methods are not only computationally demanding but also require manual adjustments; thus, a fast and reliable solution is still lacking. To address these challenges, we introduce a stitching method that leverages the advantages of crack image-segmentation models. This method first utilizes the Mask R-CNN model for the identification of crack regions as regions of interest (ROIs) within images. These regions are then used to calculate keypoints of the scale-invariant feature transform (SIFT), and descriptors for these keypoints are computed with the original images for image matching and stitching. Compared with traditional methods, our approach significantly reduces the computational time; by 98.6% in comparison to the Brute Force (BF) matcher, and by 58.7% with respect to the Fast Library for Approximate Nearest Neighbors (FLANN) matcher. Our stitching results on images with different degrees of overlap or changes in shooting posture show superior structural similarity index (SSIM) values, demonstrating excellent detail-matching performance. Moreover, the ability to measure complete crack images is indicated by the relative error of 7%, which is significantly better than that of traditional methods.

COMPARING ROTATION-ROBUST MECHANISMS IN LOCAL FEATURE MATCHING: HAND-CRAFTED VS. DEEP LEARNING ALGORITHMS

The objective of this research is to conduct a performance comparison between hand-crafted feature matching algorithms and deep learning-based counterparts in the context of rotational variances. Hand-crafted algorithms underwent testing utilizing FLANN (Fast Library for Approximate Nearest Neighbors) as the matcher and RANSAC (Random sample consensus) for outlier detection and elimination, contributing to enhanced accuracy in the results. Surprisingly, experiments revealed that hand-crafted algorithms could yield comparable or superior results to deep learning-based algorithms when exposed to rotational variances. Notably, the application of horizontally flipped images showcased a distinct advantage for deep learning-based algorithms, demonstrating significantly improved results compared to their hand-crafted counterparts. While deep learning-based algorithms exhibit technological advancements, the study found that hand-crafted algorithms like AKAZE and AKAZE-SIFT could effectively compete with their deep learning counterparts, particularly in scenarios involving rotational variances. However, the same level of competitiveness was not observed in horizontally flipped cases, where hand-crafted algorithms exhibited suboptimal results. Conversely, deep learning algorithms such as DELF demonstrated superior results and accuracy in horizontally flipped scenarios. The research underscores that the choice between hand-crafted and deep learning-based algorithms depends on the specific use case. Hand-crafted algorithms exhibit competitiveness, especially in addressing rotational variances, while deep learning-based algorithms, exemplified by DELF, excel in scenarios involving horizontally flipped images, showcasing the unique advantages each approach holds in different contexts.

A Fast Image Matching Method Based on Improved SURF

In order to solve the problems of low matching accuracy, slow speed and high system overhead in image matching methods, a rotation binary descriptor construction method based on Speed Up Robust Features (SURF) feature point detection is designed by using different Fast Library for Approximate Nearest Neighbors (FLANN) parameters and the filtering mechanism to screen out wrong matches according to the types of feature descriptors constructed in different feature extraction algorithms. This method ensures scale and rotation invariant while simplifying the representation of feature descriptors and speeding up the calculation speed in the initial stage of matching by combining the binary characteristics of descriptors. Finally, the Hamming distance is used as the filtering mechanism to improve the success rate of the final matching. The experimental results show that the accuracy of image matching is improved by 1.5% and the matching time is improved by 0.116s, while the robustness of the image to noise and rotation is ensured.

A global optimization generation method of stitching dental panorama with anti-perspective transformation.

To address the limitation of narrow field-of-view in local oral cavity images that fail to capture large-area targets at once, this paper designs a method for generating natural dental panoramas based on oral endoscopic imaging that consists of two main stages: the anti-perspective transformation feature extraction and the coarse-to-fine global optimization matching. In the first stage, we increase the number of matched pairs and improve the robustness of the algorithm to viewpoint transformation by normalizing the anti-affine transformation region extracted from the Gaussian scale space and using log-polar coordinates to compute the gradient histogram of the octagonal region to obtain the set of perspective transformation resistant feature points. In the second stage, we design a coarse-to-fine global optimization matching strategy. Initially, we incorporate motion smoothing constraints and improve the Fast Library for Approximate Nearest Neighbors (FLANN) algorithm by utilizing neighborhood information for coarse matching. Then, we eliminate mismatches via homography-guided Random Sample Consensus (RANSAC) and further refine the matching using the Levenberg-Marquardt (L-M) algorithm to reduce cumulative errors and achieve global optimization. Finally, multi-band blending is used to eliminate the ghosting due to unalignment and make the image transition more natural. Experiments show that the visual effect of dental panoramas generated by the proposed method is significantly better than that of other methods, addressing the problems of sparse splicing discontinuities caused by sparse keypoints, ghosting due to parallax, and distortion caused by the accumulation of errors in multi-image splicing in oral endoscopic image stitching.

Diagnosis of Pulmonary Nodules on CT Images Using YOLOv4

In this paper, the Scale-Invariant Feature Transform (SIFT) and Fast Library for Approximate Nearest Neighbors (FLANN) based algorithm is used to detect the abnormalities in the National Lung Screening Trial (NLST) CT scans as the exact clinical nodule locations are not provided in the dataset. These identified nodules on NLST CT Scans are then annotated using LabelImg tool. This process consumes time and so furthermore, the automatic nodule detection, You Only Look Once version 4 (YOLOv4) object detection model is implemented. The YOLOv4 object detection model is provided with total of 4187 labelled images in form of training (70%), validation (20%), and test (10%) datasets. Our YOLOv4 model achieves precision of 95%, sensitivity of 81% and mean Average Precision (mAP) of 89.1%.

Using machine learning and computer vision to estimate the angular velocity of wind turbines in smart grids remotely

Today, power generation from clean and renewable resources such as wind and solar is of great salience. Smart grid technology efficiently responds to the increasing demand for electric power. Intelligent monitoring, control, and maintenance of wind energy facilities are indispensable to increase the performance and efficiency of smart grids (SGs). Integration of state-of-the-art machine learning algorithms and vision sensor networks approaches pave the way toward enhancing the wind farms’ performance. The generating power in a wind turbine farm is the most critical parameter that should be measured accurately. Produced power is highly related to weather patterns, and a new farm in a near area is also likely to have similar energy generation. Therefore, accurate and perpetual prediction models of the existing wind farms can be led to develop new stations with lower costs. The paper aims to estimate the angular velocity of turbine blades using vision sensors and signal processing. The high wind in the wind farm can cause the camera to vibrate in successive frames, and the noise in the input images can also strengthen the problem. Thanks to couples of solid computer vision algorithms, including FAST (Features from Accelerated Segment Test), SIFT (Scale-Invariant Feature Transform), SURF (Speeded Up Robust Features), BF (Brute-Force), FLANN (Fast Library for Approximate Nearest Neighbors), AE (Autoencoder), and SVM (support vector machines), this paper accurately localizes the Hub and track the presence of the Blade in consecutive frames of a video stream. The simulation results show that determining the hub location and the blade presence in sequential frames results in an accurate estimation of wind turbine angular velocity with 95.36% accuracy.

Combination of SIFT and Canny Edge Detection for Registration Between SAR and Optical Images

Scale-invariant feature transform (SIFT) has been successfully used for optical image registration, but it cannot produce satisfying results when directly applied to synthetic aperture radar (SAR) images. In the present study, a novel method is proposed for registration between SAR and optical images. First, candidate keypoints are detected using SIFT algorithm. Then, Canny edge detection algorithm is adopted to remove the edge points which are wrong candidate points. Next, SIFT descriptors are generated from these correct keypoints. Last, the fast library for approximate nearest neighbors (FLANNs) algorithm is applied to search matching points in high-dimensional space. Experimental results show that the proposed approach is significantly more accurate and much faster than the original SIFT algorithm.

Dog Identification Method Based on Muzzle Pattern Image

Currently, invasive and external radio frequency identification (RFID) devices and pet tags are widely used for dog identification. However, social problems such as abandoning and losing dogs are constantly increasing. A more effective alternative to the existing identification method is required and the biometrics can be the alternative. This paper proposes an effective dog muzzle recognition method to identify individual dogs. The proposed method consists of preprocessing, feature extraction, matching, and postprocessing. For preprocessing, proposed resize and histogram equalization are used. For feature extraction algorithm, Scale Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF), Binary Robust Invariant Scaling Keypoints (BRISK) and Oriented FAST, and Rotated BRIEF (ORB) are applied and compared. For matching, Fast Library for Approximate Nearest Neighbors (FLANN) is used for SIFT and SURF, and hamming distance are used for BRISK and ORB. For postprocessing, two techniques to reduce incorrect matches are proposed. The proposed method was evaluated with 55 dog muzzle pattern images acquired from 11 dogs and 990 images augmented by the image deformation (i.e., angle, illumination, noise, affine transform). The best Equal Error Rate (EER) of the proposed method was 0.35%, and ORB was the most appropriate for the dog muzzle pattern recognition.

Wind turbine tower detection using feature descriptors and deep learning

Wind Turbine Towers (WTTs) are the main structures of wind farms. They are costly devices that must be thoroughly inspected according to maintenance plans. Today, existence of machine vision techniques along with unmanned aerial vehicles (UAVs) enable fast, easy, and intelligent visual inspection of the structures. Our work is aimed towards developing a vision-based system to perform Nondestructive tests (NDTs) for wind turbines using UAVs. In order to navigate the flying machine toward the wind turbine tower and reliably land on it, the exact position of the wind turbine and its tower must be detected. We employ several strong computer vision approaches such as Scale-Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF), Features from Accelerated Segment Test (FAST), Brute-Force, Fast Library for Approximate Nearest Neighbors (FLANN) to detect the WTT. Then, in order to increase the reliability of the system, we apply the ResNet, MobileNet, ShuffleNet, EffNet, and SqueezeNet pre-trained classifiers in order to verify whether a detected object is indeed a turbine tower or not. This intelligent monitoring system has auto navigation ability and can be used for future goals including intelligent fault diagnosis and maintenance purposes. The simulation results show the accuracy of the proposed model are 89.4% in WTT detection and 97.74% in verification (classification) problems.

An automatic approach of mapping the solar high-resolution image to Helioprojective-Cartesian coordinates system

The ground-based solar high-resolution observation usually only covers a small part of the solar disk, so determining the field of view position of an observed image on the Helioprojective-Cartesian coordinates is the first step in data processing. In the present, most solar physicists manually adjust parameters by trial and error to complete this step, because the mapping information may not be included in the flexible image transport system (FITS) header of a raw image. The manual approach cannot guarantee the accuracy. It is inefficient and has plagued many solar physicists. In this paper, we propose an automatic mapping approach by registering a high-resolution image to a standard calibrated solar full disk using image pre-processing and scale-invariant feature transform (SIFT) technology. Following steps are included. (1) Downsampling and blurring a high-resolution image to roughly fit the scale and the resolution of its corresponding full disk image; (2) scaling the gray level to 256 by local minimum and maximum to meet the requirements of applying the standard OpenCV SIFT processing package; (3) removing the limb-darkening of the reference full disk image with Cartesian to polar coordinate transform method; (4) detecting the feature points by SIFT on both high-resolution and full disk images; (5) matching corresponding point pairs by fast library for approximate nearest neighbors (FLANN); (6) calculating translation, rotation and scale parameters by random sample consensus (RANSAC); (7) recording the registration results in FITS header with standard keywords of the Solar Dynamics Observatory (SDO) image. These keywords can be automatically retrieved by the Solar SoftWare (SSW) system. With this approach, we have successfully registered both active region and quiet sun images observed by the titanium dioxide (TiO) band of the New Vacuum Solar Telescope (NVST) to the continuum of the helioseismic and magnetic imager (HMI) instrument on the SDO, and Hα active region images of NVST to the Global Oscillation Network Group (GONG) with high accuracy (0.25 arcsec for photosphere and 1 arcsec for chromosphere). If a high-resolution image shows bright structures, such as flares, the Hα image of NVST could be registered to 304 A image of the atmospheric imaging assembly (AIA) on SDO with the accuracy of 1 arcsec as well. In addition, the images observed by Hα blue/red wings (±0.7 A) of NVST, TiO of the Goode solar telescope (GST), Hα of the Optical and Near-infrared Solar Eruption Tracer (ONSET) are also successfully registered to SDO/HMI continuum or SDO/AIA 304 A full disk image. Iterative processing is applied to improve accuracy. An automatic SDO or GONG image downloading procedure and an extra interactive user interface are also integrated. This approach has met the requirement of mapping a high-resolution image to the Helioprojective-Cartesian coordinates of full disk. Its weakness is that if a high-resolution image does not show high contrast, such as limb observation of photosphere or quiet sun of chromosphere, it is difficult to get a mapping result, because the whole strategy is based on image feature detection. The procedure will be merged into the observation system of NVST in the future, and will provide great convenience for the solar physicists to use the high-resolution observation data.

A hybrid approach to generate visually seamless aerial mosaicks from unmanned aerial vehicles

The need for understanding the terrain or conditions of large areas aerially has gained prominence as the aerial images provide a near clear coverage of the area under study. Individual image provides just a portion of the area, thus to understand the whole area, mosaicking or stitching of these i mages is needed. Image mosaicking aids in providing with a ”Big Picture” as an outcome by joining the images taken during the flight. In this paper we propose a method which aims at generating a seamless aerial mosaick using only the images captured by the UAV as input. This involves identifying candidate images from the images captured by the UAV periodically during its flight and stitching the images together. This method evaluates various feature descriptors and feature matching techniques that can be integrated into the mosaicking system. The proposed work is a hybrid approach that uses the Scale Invariant Feature Transform (SIFT) for feature extraction and the key features are matched using the Fast Library for Approximate Nearest Neighbors (FLANN). RANdom Sample Consensus (RANSAC), is used for the removal of features that are redundant or act as outliner, providing candidates for Homography estimation. This is followed by image stitching that involves the use of Multi-Band Blending to produce a visually seamless mosaick. The results obtained were evaluated for quality using Universal Quality index Measure (QIM) and is found to be perfect.

Development and Experimental Evaluation of a 3D Vision System for Grinding Robot.

If the grinding robot can automatically position and measure the machining target on the workpiece, it will significantly improve its machining efficiency and intelligence level. However, unfortunately, the current grinding robot cannot do this because of economic and precision reasons. This paper proposes a 3D vision system mounted on the robot’s fourth joint, which is used to detect the machining target of the grinding robot. Also, the hardware architecture and data processing method of the 3D vision system is described in detail. In the data processing process, we first use the voxel grid filter to preprocess the point cloud and obtain the feature descriptor. Then use fast library for approximate nearest neighbors (FLANN) to search out the difference point cloud from the precisely registered point cloud pair and use the point cloud segmentation method proposed in this paper to extract machining path points. Finally, the detection error compensation model is used to accurately calibrate the 3D vision system to transform the machining information into the grinding robot base frame. Experimental results show that the absolute average error of repeated measurements at different locations is 0.154 mm, and the absolute measurement error of the vision system caused by compound error is usually less than 0.25 mm. The proposed 3D vision system could easily integrate into an intelligent grinding system and may be suitable for industrial sites.

Inter-frame passive-blind forgery detection for video shot based on similarity analysis

Frame insertion, deletion and duplication are common inter-frame tampering operations in digital videos. In this paper, based on similarity analysis, a passive-blind forensics scheme for video shots is proposed to detect inter-frame forgeries. This method is composed of two parts: HSV (Hue-Saturation-Value) color histogram comparison and SURF (Speeded Up Robust Features) feature extraction together with FLANN (Fast Library for Approximate Nearest Neighbors) matching for double-checking. We mainly calculate H-S and S-V color histograms of every frame in a video shot and compare the similarity between histograms to detect and locate tampered frames in the shot. Then we utilize SURF feature extraction and FLANN matching to further confirm the forgery types in the tampered locations. Experimental results demonstrate that the proposed detection method is efficient and accurate in terms of forgery identification and localization. In contrast to other inter-frame forgery detection methods, our scheme can detect three kinds of forgery operations and has its own superiority and applicability as a passive-blind detection method.

A Fast Robot Identification and Mapping Algorithm Based on Kinect Sensor.

Internet of Things (IoT) is driving innovation in an ever-growing set of application domains such as intelligent processing for autonomous robots. For an autonomous robot, one grand challenge is how to sense its surrounding environment effectively. The Simultaneous Localization and Mapping with RGB-D Kinect camera sensor on robot, called RGB-D SLAM, has been developed for this purpose but some technical challenges must be addressed. Firstly, the efficiency of the algorithm cannot satisfy real-time requirements; secondly, the accuracy of the algorithm is unacceptable. In order to address these challenges, this paper proposes a set of novel improvement methods as follows. Firstly, the ORiented Brief (ORB) method is used in feature detection and descriptor extraction. Secondly, a bidirectional Fast Library for Approximate Nearest Neighbors (FLANN) k-Nearest Neighbor (KNN) algorithm is applied to feature match. Then, the improved RANdom SAmple Consensus (RANSAC) estimation method is adopted in the motion transformation. In the meantime, high precision General Iterative Closest Points (GICP) is utilized to register a point cloud in the motion transformation optimization. To improve the accuracy of SLAM, the reduced dynamic covariance scaling (DCS) algorithm is formulated as a global optimization problem under the G2O framework. The effectiveness of the improved algorithm has been verified by testing on standard data and comparing with the ground truth obtained on Freiburg University’s datasets. The Dr Robot X80 equipped with a Kinect camera is also applied in a building corridor to verify the correctness of the improved RGB-D SLAM algorithm. With the above experiments, it can be seen that the proposed algorithm achieves higher processing speed and better accuracy.

Scalable Nearest Neighbor Algorithms for High Dimensional Data

For many computer vision and machine learning problems, large training sets are key for good performance. However, the most computationally expensive part of many computer vision and machine learning algorithms consists of finding nearest neighbor matches to high dimensional vectors that represent the training data. We propose new algorithms for approximate nearest neighbor matching and evaluate and compare them with previous algorithms. For matching high dimensional features, we find two algorithms to be the most efficient: the randomized k-d forest and a new algorithm proposed in this paper, the priority search k-means tree. We also propose a new algorithm for matching binary features by searching multiple hierarchical clustering trees and show it outperforms methods typically used in the literature. We show that the optimal nearest neighbor algorithm and its parameters depend on the data set characteristics and describe an automated configuration procedure for finding the best algorithm to search a particular data set. In order to scale to very large data sets that would otherwise not fit in the memory of a single machine, we propose a distributed nearest neighbor matching framework that can be used with any of the algorithms described in the paper. All this research has been released as an open source library called fast library for approximate nearest neighbors (FLANN), which has been incorporated into OpenCV and is now one of the most popular libraries for nearest neighbor matching.