Object Recognition Algorithm Research Articles

Kernel quaternion principal component analysis and its application in RGB-D object recognition

While the existing quaternion principal component analysis (QPCA) is a linear tool developed mainly for processing linear quaternion signals, the quaternion representation (QR) used in QPCA creates redundancy when representing a color image signal of three components by a quaternion matrix having four components. In this paper, the kernel technique is used to improve the QPCA as kernel QPCA (KQPCA) for processing nonlinear quaternion signals; in addition, both RGB information and depth information are considered to improve QR for representing RGB-D images. The improved QR fully utilizes the four-dimensional quaternion domain. We first provide the basic idea of three types of our KQPCA and then propose an algorithm for RGB-D object recognition based on bidirectional two-dimensional KQPCA (BD2DKQPCA) and the improved QR. Experimental results on four public datasets demonstrate that the proposed BD2DKQPCA-based algorithm achieves the best performance among seventeen compared algorithms including other existing PCA-based algorithms, irrespective of RGB object recognition or RGB-D object recognition. Moreover, for all compared algorithms, consideration of both RGB and depth information is shown to achieve better performance in object recognition than considering only RGB information.

Synthesis of Algorithms for Recognition of Vulnerabilities in Web Resources Using Signatures of Fuzzy Linguistic Features

The paper formalizes the problem of fuzzy recognition of vulnerabilities in web resources that is set by a complex reference description in the form of signatures of intervals of the values of fuzzy linguistic features. The reference description and the mathematics of the theory of validation of complex statistical hypotheses were applied to synthesize multi-criteria object recognition algorithms using the minimax decision rule and the Bayesian maximum of a posteriori probability and maximum likelihood criteria.

SigVox – A 3D feature matching algorithm for automatic street object recognition in mobile laser scanning point clouds

Urban road environments contain a variety of objects including different types of lamp poles and traffic signs. Its monitoring is traditionally conducted by visual inspection, which is time consuming and expensive. Mobile laser scanning (MLS) systems sample the road environment efficiently by acquiring large and accurate point clouds. This work proposes a methodology for urban road object recognition from MLS point clouds. The proposed method uses, for the first time, shape descriptors of complete objects to match repetitive objects in large point clouds. To do so, a novel 3D multi-scale shape descriptor is introduced, that is embedded in a workflow that efficiently and automatically identifies different types of lamp poles and traffic signs. The workflow starts by tiling the raw point clouds along the scanning trajectory and by identifying non-ground points. After voxelization of the non-ground points, connected voxels are clustered to form candidate objects. For automatic recognition of lamp poles and street signs, a 3D significant eigenvector based shape descriptor using voxels (SigVox) is introduced. The 3D SigVox descriptor is constructed by first subdividing the points with an octree into several levels. Next, significant eigenvectors of the points in each voxel are determined by principal component analysis (PCA) and mapped onto the appropriate triangle of a sphere approximating icosahedron. This step is repeated for different scales. By determining the similarity of 3D SigVox descriptors between candidate point clusters and training objects, street furniture is automatically identified. The feasibility and quality of the proposed method is verified on two point clouds obtained in opposite direction of a stretch of road of 4km. 6 types of lamp pole and 4 types of road sign were selected as objects of interest. Ground truth validation showed that the overall accuracy of the ∼170 automatically recognized objects is approximately 95%. The results demonstrate that the proposed method is able to recognize street furniture in a practical scenario. Remaining difficult cases are touching objects, like a lamp pole close to a tree.

SpatialVISION – software module for objects identification before robotic gripping

In this paper we present a software module for object recognition before robotic gripping with an anthropomorphic gripper. To this effect, recognizing the shape of an object using digital image processing algorithms, falling in the area of digital sequences processing, is important. This area has evolved a lot lately, in particular due to the progress of computing resources, involving parallel processing of each basic unit of an image. The paper summarizes the theoretical background and the results related to the software developed for this purpose, called SpatialVISION. To implement the SpatialVISION application, the implementation of the SIFT algorithm was used, which is a personal approach in C #, using EmguCV library for working with images. To use the SpatialVISION application, a tool called SVAdminUtility was also created, which is used to populate a database with three-dimensional objects that will later be recognized. In structural terms, the SVAdminUtility application has four functional components namely: SVAdminUtility component; Collada component; Viewer component and XSD2NET component. To assess the SpatialVISION application, an experiment was conducted, whose results are presented in the paper, which consists in measuring the average time of execution of the object recognition algorithm using two types of images captured by two types of cameras: a WEB camera and a Canon PowerShot camera.

A hardware-oriented histogram of oriented gradients algorithm and its VLSI implementation

A challenging and important issue for object recognition is feature extraction on embedded systems. We report a hardware implementation of the histogram of oriented gradients (HOG) algorithm for real-time object recognition, which is known to provide high efficiency and accuracy. The developed hardware-oriented algorithm exploits the cell-based scan strategy which enables image-sensor synchronization and extraction-speed acceleration. Furthermore, buffers for image frames or integral images are avoided. An image-size scalable hardware architecture with an effective bin-decoder and a parallelized voting element (PVE) is developed and used to verify the hardware-oriented HOG implementation with the application of human detection. The fabricated test chip in 180 nm CMOS technology achieves fast processing speed and large flexibility for different image resolutions with substantially reduced hardware cost and energy consumption.

Digital pathology in immuno-oncology – a roadmap for clinical development

ABSTRACTIntroduction: Immuno-oncology clinical development requires the sophisticated understanding of intersecting immune and oncology pathways. These challenges are increasingly being addressed with digital solutions combining high-resolution images of histological specimens with complex object-recognition algorithms. Despite these advancements, the inertia of migrating to digital images remains significant to many parties.Areas covered: We provide a roadmap to implement digital pathology in immuno-oncology clinical development. We discuss its slow adoption rate and the controversies of digital images versus glass-slide reads. Large hurdles include the lack of implementable validation and quality systems that meet regulatory criteria while balancing the rapid changes of immuno-oncology development and technology innovation. We include a practical validation framework and provide examples of immuno-oncology-relevant applications that benefit from digital applications.Expert commentary: Multi-platform biomarker models that assess the tumor microenvironment already integrate digital pathology and image analyses. Our ability to increase user confidence and provide robust documentation that withstands regulatory and clinical scrutiny is essential to leverage these digital pathology technologies for decision-making purposes. Moreover, implementing validation and quality assurance activities for digital pathology in oncology clinical development will play a vital role to establish standards and use cases, which may then serve as subsequent metrics for clinical consideration.

Autonomous metrology for robot mounted 3D vision systems

Using a metrology system simulation approach, an algorithm is presented to determine the best position for a robot mounted 3D vision system. Point cloud data is simulated, taking into account sensor performance, to create a ranked list of the best camera positions. These can be used by a robot to autonomously determine the most advantageous camera position for locating a target object. The algorithm is applied to an Ensenso active stereo 3D camera. Results show that when used in combination with a RANSAC object recognition algorithm, it increased positional precision by two orders of magnitude, from worst to best case.

The 3D object recognition with environmental adaptability based on VFH descriptor and region growing segmentation

3D object recognition is a basic research in the machine vision field. Microsoft KINECT V2 is utilized to collect external environmental information. The point cloud file is obtained after processing the collected information. In order to filter the point cloud and obtain point cloud model of a single object in the environment after region growing segmentation, the point cloud is applied to point cloud library. Then, the VFH descriptor of the point cloud model is calculated. After inputting point cloud model of the trained target, the point cloud model with the minimum CHI square distance between the VFH descriptor of the target and VFH descriptor of the point cloud model can be found. The 3D object corresponding to the found model is the identified object. For the 3D object recognition in an unfamiliar environment, the algorithm of 3D object recognition with environmental adaptability is proposed. After the 3D object recognition system built, the physical verification is conducted about the proposed algorithm. Giving the target model, the system successfully identifies the 3D object in the unfamiliar environment, that demonstrates the correctness of the algorithm.

Sub-OBB based object recognition and localization algorithm using range images

This paper presents a novel approach to recognize and estimate pose of the 3D objects in cluttered range images. The key technical breakthrough of the developed approach can enable robust object recognition and localization under undesirable condition such as environmental illumination variation as well as optical occlusion to viewing the object partially. First, the acquired point clouds are segmented into individual object point clouds based on the developed 3D object segmentation for randomly stacked objects. Second, an efficient shape-matching algorithm called Sub-OBB based object recognition by using the proposed oriented bounding box (OBB) regional area-based descriptor is performed to reliably recognize the object. Then, the 3D position and orientation of the object can be roughly estimated by aligning the OBB of segmented object point cloud with OBB of matched point cloud in a database generated from CAD model and 3D virtual camera. To detect accurate pose of the object, the iterative closest point (ICP) algorithm is used to match the object model with the segmented point clouds. From the feasibility test of several scenarios, the developed approach is verified to be feasible for object pose recognition and localization.

Webly-Supervised Fine-Grained Visual Categorization via Deep Domain Adaptation.

Learning visual representations from web data has recently attracted attention for object recognition. Previous studies have mainly focused on overcoming label noise and data bias and have shown promising results by learning directly from web data. However, we argue that it might be better to transfer knowledge from existing human labeling resources to improve performance at nearly no additional cost. In this paper, we propose a new semi-supervised method for learning via web data. Our method has the unique design of exploiting strong supervision, i.e., in addition to standard image-level labels, our method also utilizes detailed annotations including object bounding boxes and part landmarks. By transferring as much knowledge as possible from existing strongly supervised datasets to weakly supervised web images, our method can benefit from sophisticated object recognition algorithms and overcome several typical problems found in webly-supervised learning. We consider the problem of fine-grained visual categorization, in which existing training resources are scarce, as our main research objective. Comprehensive experimentation and extensive analysis demonstrate encouraging performance of the proposed approach, which, at the same time, delivers a new pipeline for fine-grained visual categorization that is likely to be highly effective for real-world applications.

Object Recognition using Multicore Processor

Object recognition is the task of finding an object from a given image. A human being can recognize objects that are in a video sample or an image with different size, position or viewpoint. This ability is becoming a common and developing subject of research. Objective: The main aim of this paper is to implement such human ability as an object recognition system onto a computer and address the issues related to scaling, rotation invariance, and computational complexity and to recognize an object in a cluttered background. Methods: In this paper, object recognition is performed based on attention based object recognition algorithm. Gaussian and Difference of Gaussian filters are used for image pre-processing. Saliency map is generated, and keypoints are extracted using Laplace of Gaussian approach. Then keypoint descriptor vectors are generated, and matching of test image and training image is performed using a fast nearest neighbour algorithm. Findings: The scaling factor of 1.2 is applied to increase the detection rate under differing image rotations. Implementation is done using parallelization concept in a multicore processor using OpenMP programming model. Improvement: The performance of serial and parallel execution of object recognition is analyzed, and the performance can still be increased by using OpenMP directives such as sections and tasks.

Object Detection and Recognition Algorithm of Moving UAV Based on Machine Vision

To conquer disadvantages of slow speed of target tracking algorithm in original distribution field as well as easiness of being caught in local optimal solution, one target tracking algorithm of real-time distribution field based on global matching is presented in the Thesis, thus remarkably improving performance of target tracking algorithm in distribution field. In proposed algorithm, relevant correlation coefficients will be used to substitute the similarity between target distribution filed of original L1 norm measurement and candidate distribution field. As a consequence, the target search process can concert from time domain operation to operation processing of frequency domain among which the latter has lower computation complexity and capability of global search of target position so as to conquer disadvantages such as randomness caused by sparse sampling and that the gradient descent of target tracking algorithm in original distribution field is liable to be caught in local optimal solution. In 12 challenging video sequences, compared with multiple-instance learning and tracking algorithm and tracking algorithm of original distribution field, the method proposed in the Thesis has acquired the optimum performance in tracking accuracy, success rate and speed.

Reflective workpiece detection and localization for flexible robotic cells

A smart vision system for industrial robotic cells is presented. It can recognize and localize a reflective workpiece, and allows for automatic adjustments of the robot program. The purpose of the study is to improve industrial robots awareness of the environment and to increase adaptability of the manufacturing processes where full control over environment is not achievable. This approach is particularly relevant to small batch robotic production, often suffering from only partial control of the process parameters, such as the order of jobs, workpiece position, or illumination conditions.A distinguishing aspect of the study is detection of workpieces made of diverse materials, including shiny metals. Reflective surfaces are common in the industrial manufacturing, but are rarely considered in the research on object recognition because they hinder many of the object recognition algorithms. The proposed solution has been qualified and tested on a selected benchmark in a realistic workshop environment with artificial light conditions. The training of the object recognition software is an automatic process and can be executed by non-expert industrial users to allow for recognition of different types of objects. HighlightsA machine vision system for detection of workpieces in a robotic cell is proposed.The system can recognize objects with specular surfaces (made of reflective metal).It estimates the location of the object on the workbench from a single camera image.It adjusts the robot program according to the presence of the workpiece and its type.

Genetic optical design for a compressive sensing task

We present a sophisticated optical design method for reducing the number of photodetectors for a specific sensing task. The chosen design parameter is the point spread function, and the selected task is object recognition. The point spread function is optimized iteratively with a genetic algorithm for object recognition based on a neural network. In the experimental demonstration, binary classification of face and non-face datasets was performed with a single measurement using two photodetectors. A spatial light modulator operating in the amplitude modulation mode was provided in the imaging optics and was used to modulate the point spread function. In each generation of the genetic algorithm, the classification accuracy with a pattern displayed on the spatial light modulator was fed-back to the next generation to find better patterns. The proposed method increased the accuracy by about 30 % compared with a conventional imaging system in which the point spread function was the delta function. This approach is practically useful for compressing the cost, size, and observation time of optical sensors in specific applications, and robust for imperfections in optical elements.

Object Occlusion Detection Using Automatic Camera Calibration for a Wide-Area Video Surveillance System

This paper presents an object occlusion detection algorithm using object depth information that is estimated by automatic camera calibration. The object occlusion problem is a major factor to degrade the performance of object tracking and recognition. To detect an object occlusion, the proposed algorithm consists of three steps: (i) automatic camera calibration using both moving objects and a background structure; (ii) object depth estimation; and (iii) detection of occluded regions. The proposed algorithm estimates the depth of the object without extra sensors but with a generic red, green and blue (RGB) camera. As a result, the proposed algorithm can be applied to improve the performance of object tracking and object recognition algorithms for video surveillance systems.

EXPLORING REGULARITIES FOR IMPROVING FAÇADE RECONSTRUCTION FROM POINT CLOUDS

Abstract. (Semi)-automatic facade reconstruction from terrestrial LiDAR point clouds is often affected by both quality of point cloud itself and imperfectness of object recognition algorithms. In this paper, we employ regularities, which exist on façades, to mitigate these problems. For example, doors, windows and balconies often have orthogonal and parallel boundaries. Many windows are constructed with the same shape. They may be arranged at the same lines and distance intervals, so do different windows. By identifying regularities among objects with relatively poor quality, these can be applied to calibrate the objects and improve their quality. The paper focuses on the regularities among the windows, which is the majority of objects on the wall. Regularities are classified into three categories: within an individual window, among similar windows and among different windows. Nine cases are specified as a reference for exploration. A hierarchical clustering method is employed to identify and apply regularities in a feature space, where regularities can be identified from clusters. To find the corresponding features in the nine cases of regularities, two phases are distinguished for similar and different windows. In the first phase, ICP (iterative closest points) is used to identify groups of similar windows. The registered points and a number of transformation matrices are used to identify and apply regularities among similar windows. In the second phase, features are extracted from the boundaries of the different windows. When applying regularities by relocating windows, the connections, called chains, established among the similar windows in the first phase are preserved. To test the performance of the algorithms, two datasets from terrestrial LiDAR point clouds are used. Both show good effects on the reconstructed model, while still matching with original point cloud, preventing over or under-regularization.

EXPLORING REGULARITIES FOR IMPROVING FAÇADE RECONSTRUCTION FROM POINT CLOUDS

(Semi)-automatic facade reconstruction from terrestrial LiDAR point clouds is often affected by both quality of point cloud itself and imperfectness of object recognition algorithms. In this paper, we employ regularities, which exist on façades, to mitigate these problems. For example, doors, windows and balconies often have orthogonal and parallel boundaries. Many windows are constructed with the same shape. They may be arranged at the same lines and distance intervals, so do different windows. By identifying regularities among objects with relatively poor quality, these can be applied to calibrate the objects and improve their quality. The paper focuses on the regularities among the windows, which is the majority of objects on the wall. Regularities are classified into three categories: within an individual window, among similar windows and among different windows. Nine cases are specified as a reference for exploration. A hierarchical clustering method is employed to identify and apply regularities in a feature space, where regularities can be identified from clusters. To find the corresponding features in the nine cases of regularities, two phases are distinguished for similar and different windows. In the first phase, ICP (iterative closest points) is used to identify groups of similar windows. The registered points and a number of transformation matrices are used to identify and apply regularities among similar windows. In the second phase, features are extracted from the boundaries of the different windows. When applying regularities by relocating windows, the connections, called chains, established among the similar windows in the first phase are preserved. To test the performance of the algorithms, two datasets from terrestrial LiDAR point clouds are used. Both show good effects on the reconstructed model, while still matching with original point cloud, preventing over or under-regularization.

Object recognition algorithm for the automatic identification and removal of invasive fish

Invasive fish species are a growing threat worldwide, causing great harm to biodiversity and ecosystems, and leading to large economic losses. As the most introduced group of aquatic animals in the world, fish are also one of the most threatened. For species that are considered invasive, removing them is the best way to reduce the long-term cost of eradication or control. This paper proposes an object recognition algorithm to automatically identify fish species. Our previous work on general object recognition, called Evolution-COnstructed (ECO) Features, was modified and adapted to construct features and use AdaBoost to classify different fish species. The proposed algorithm does not depend on human experts to design features for fish species classification, but constructs efficient features automatically. Results from experiments show the proposed method obtained an average of 98.9% classification accuracy with a standard deviation of 0.96% with a dataset composed of 8 fish species and a total of 1049 images. Using this algorithm, a fish monitoring system can be built to remove invasive species and monitor native fish abundance, distribution, and size with minimal collateral impact and fish suffering.

Multi-sensor 3D object dataset for object recognition with full pose estimation

In this work, we propose a new dataset for 3D object recognition using the new high-resolution Kinect V2 sensor and some other popular low-cost devices like PrimeSense Carmine. Since most already existing datasets for 3D object recognition lack some features such as 3D pose information about objects in the scene, per pixel segmentation or level of occlusion, we propose a new one combining all this information in a single dataset that can be used to validate existing and new 3D object recognition algorithms. Moreover, with the advent of the new Kinect V2 sensor we are able to provide high-resolution data for RGB and depth information using a single sensor, whereas other datasets had to combine multiple sensors. In addition, we will also provide semiautomatic segmentation and semantic labels about the different parts of the objects so that the dataset could be used for testing robot grasping and scene labeling systems as well as for object recognition.

Metric learning based object recognition and retrieval

Object recognition and retrieval is an important topic in intelligent robotics and pattern recognition, where an effective recognition engine plays an important role. To achieve a good performance, we propose a metric learning based object recognition algorithm. To represent the invariant object features, including local shape details and global body parts, a novel multi-scale invariant descriptor is proposed. Different types of invariant features are represented in multiple scales, which makes the following metric learning algorithm effective. To reduce the effect of noise and improve the computing efficiency, an adaptive discrete contour evolution method is also proposed to extract the salient feature points of object. The recognition algorithm is explored based on metric learning method and the object features are summarized as histograms inspired from the Bag of Words (BoW). The metric learning methods are employed to learn object features according to their scales. The proposed method is invariant to rotation, scale variation, intra-class variation, articulated deformation and partial occlusion. The recognition process is fast and robust for noise. This method is evaluated on multiple benchmark datasets and the comparable experimental results indicate the effectiveness of our method.