HOG Feature Descriptor Research Articles

Two-Stage Traffic Sign Classification System

Traffic Sign Recognition (TSR) has received widespread attention due to increased traffic accidents caused by a failure to recognize road traffic signs. Most Traffic Sign Recognition remains challenging due to the illumination of natural scenes, and the diversity in traffic signs [31]. To solve the problem, we proposed a two-stage classification system to overcome the above-mentioned challenges and classify the traffic sign boards efficiently. During the first phase, colour based histogram equalizations applied to eliminates the illumination factors and then traffic sign boards are classified into many shapes, which includes circular, triangular, hexagonal, and diamond traffic signs boards based on the modified HOG feature descriptors and various machine learning classifiers are compared [12]. In the second phase, circular and triangular traffic signs are further categorized into specific subclasses using circular and triangular signs trained Convolutional Neural Network (CNN) respectively to overcome intra-class variation among the similar sign board. Experimentation was conducted on German Traffic Sign Recognition Benchmark dataset (GTSRB) [30] and the results show 99.96% top-1 accuracy during the first stage of classification. In the second stage, there is accuracy of 99.40% for circular traffic signs and 97.50% for triangular traffic signs.

ROS-Industrial based robotic cell for Industry 4.0: Eye-in-hand stereo camera and visual servoing for flexible, fast, and accurate picking and hooking in the production line

The fourth industrial revolution envisages the use of modern smart technologies to automate traditional manufacturing and industrial practices. However, industrial robots execute mostly pre-programmed jobs and are not able to face challenging tasks in unstructured environments. Industry 4.0 pushes for flexibility on target changes and autonomy. In line with the new principles of Industry 4.0, the proposed work describes an autonomous industrial cell that employs several smart technologies for loading jewelry pieces from a conveyor belt to a hooking frame built on purpose. The cell involves an industrial robot, a custom gripper, pneumatic and electric actuators with the aim of moving and opening the frame hooks, and a custom vision pipeline for detecting the feature of interest during the picking and hooking phases. The implemented pipeline makes use of a stereo camera pair mounted under the robot gripper and two fixed monocular cameras. The method employs HOG feature descriptors and machine learning algorithms for the detection. The software architecture is a component-based designed architecture that uses ROS as the underlying framework and ROS-Industrial packages to control the robot. The robot is controlled with position-based commands to reach intermediate positions in the workspace and with velocity command to implement a visual servoing control scheme that runs at 30 Hz and adjusts the robot position with the feedback of the vision during picking and hooking. The proposed visual servoing approach, thanks to the design of the stereo camera and choice of the optics, is able to perceive the features until the final movement phase, differently from most of the visual servoing employed in the literature that, due to the use of RGB-D camera or other vision apparatus, use an open control loop at a standoff distance. The presented work reaches an accuracy of 95% with a cycle time under 8 s.

Target Detection in Video Images Using HOG-Based Cascade Classifier

Detecting small objects using computer vision is a challenging task due to their small size in the image and therefore the lack of features when describing them. In this paper, a computer was trained to detect three small balls using 20 levels of the AdaBoost cascade classifier. The features of the balls in each level are described using the HOG feature descriptor. Three balls were recorded in practice at various distances (d = 2, 3, 4, ..., 10 m) from the camera and the targets (balls). The frames are then taken from the videos and resized using five magnification factors (RS = 1, 3, 5, 7, and 9) to make the balls seem as they should. According to the results, the detection rate of balls at all distances was 80% when using the magnification factor RS = 1, 90% when using the magnification factor RS = 3, 5, and 7, and 100% when using the magnification factor RS = 9. The suggested approach was also used in calculating the height and width of the detected balls. The overall results indicated that the height and width of the balls dwindle as the distance between the camera and the targets increases.

A novel multi-atlas segmentation approach under the semi-supervised learning framework: Application to knee cartilage segmentation

Background and objective: Multi-atlas based segmentation techniques, which rely on an atlas library comprised of training images labeled by an expert, have proven their effectiveness in multiple automatic segmentation applications. However, the usage of exhaustive patch libraries combined with the voxel-wise labeling incur a large computational cost in terms of memory requirements and execution times. Methods: To confront this shortcoming, we propose a novel two-stage multi-atlas approach designed under the Semi-Supervised Learning (SSL) framework. The main properties of our method are as follows: First, instead of the voxel-wise labeling approach, the labeling of target voxels is accomplished here by exploiting the spectral content of globally sampled datasets from the target image, along with their spatially correspondent data collected from the atlases. Following SSL, voxels classification is boosted by incorporating unlabeled data from the target image, in addition to the labeled ones from atlas library. Our scheme integrates constructively fruitful concepts, including sparse reconstructions of voxels from linear neighborhoods, HOG feature descriptors of patches/regions, and label propagation via sparse graph constructions. Segmentation of the target image is carried out in two stages: stage-1 focuses on the sampling and labeling of global data, while stage-2 undertakes the above tasks for the out-of-sample data. Finally, we propose different graph-based methods for the labeling of global data, while these methods are extended to deal with the out-of-sample voxels. Results: A thorough experimental investigation is conducted on 76 subjects provided by the publicly accessible Osteoarthritis Initiative (OAI) repository. Comparative results and statistical analysis demonstrate that the suggested methodology exhibits superior segmentation performance compared to the existing patch-based methods, across all evaluation metrics (DSC:88.89%, Precision: 89.86%, Recall: 88.12%), while at the same time it requires a considerably reduced computational load (>70% reduction on average execution time with respect to other patch-based). In addition, our approach is favorably compared against other non patch-based and deep learning methods in terms of performance accuracy (on the 3-class problem). A final experimentation on a 5-class setting of the problems demonstrates that our approach is capable of achieving performance comparable to existing state-of-the-art knee cartilage segmentation methods (DSC:88.22% and DSC:85.84% for femoral and tibial cartilage respectively).

Non- Destructive and Cost-Effective Silkworm Cocoon Gender Classification with Hog Feature Descriptors

Non- Destructive and Cost-Effective Silkworm Cocoon Gender Classification with Hog Feature Descriptors

Robust Real Time Multiple Human Detection and Tracking for Automatic Visual Surveillance System

Detection of human for visual surveillance system provides most important rule for advancement in the design of future automation systems. Human detection and tracking are important for future automatic visual surveillance system (AVSS). In this paper we have proposed a flexible technique for proper human detection and tracking for the design of AVSS. We used graph cut for segment human as a foreground image by eliminating background, extract some feature points by using HOG, SVM classifier for proper classification and finally we used particle filter for tracking those of detected human. Our system can easily detect and track humans in poor lightening conditions, color, size, shape, and clothing due to the use of HOG feature descriptor and particle filter. We use graph cut based segmentation technique, therefore our system can handle occlusion at about 88%. Due to the use of HOG to extract features our system can properly work in indoor as well as outdoor environments with 97.61% automatic human detection and 92% automatic human detection and tracking accuracy of multiple human

A hybrid fusion framework to multi-modal bio metric identification

In the recent decade, comprehensive research efforts have been carried out as the promising modality of bio metrics on humans’ physical features for person recognition. Despite this, the main issue encountered in identifying individuals is obtaining rich representation for multi-modal data that is invariant to diverse physical traits. The shortcomings of uni-modal bio metric systems can be tackled by combining derived knowledge from several modalities of bio metric systems embedded with several physical characteristics like Ear, Face, Iris, and Gait. This paper proposes a novel multi-modal bio metric identification framework based on a hybrid multi-phase feature fusion to render compact knowledge from multiple model traits. We employed transfer learning through several pertained networks such as Resnet101, Resnet-Inceptionv2, Densenet201, AlexNet, and Inceptionv2 to fuse with handcrafted feature vectors extracted via Hog feature descriptor. The fusion is performed using Discriminant Correlation Analysis (DCA) and Canonical Correlation Analysis (CCA) at each single and hybrid phase. Three state of the art bio metric databases, namely Face, Gait, and Ear, was utilized to evaluate the proposed framework. The proposed framework based on multi-phase hybrid fusion achieved up to 96.6% of identification accuracy using multi traits. Experimental results confirm the superior results over other recent multi-modal bio metric variants.

A New Pedestrian Feature Description Method Named Neighborhood Descriptor of Oriented Gradients

In view of the issue of the mechanism for enhancing the neighbourhood relationship of blocks of HOG, this paper proposes neighborhood descriptor of oriented gradients (NDOG), an improved feature descriptor based on HOG, for pedestrian detection. To obtain the NDOG feature vector, the algorithm calculates the local weight vector of the HOG feature descriptor, while integrating spatial correlation among blocks, concatenates this weight vector to the tail of the HOG feature descriptor, and uses the gradient norm to normalize this new feature vector. With the proposed NDOG feature vector along with a linear SVM classifier, this paper develops a complete pedestrian detection approach. Experimental results for the INRIA, Caltech-USA, and ETH pedestrian datasets show that the approach achieves a lower miss rate and a higher average precision compared with HOG and other advanced methods for pedestrian detection especially in the case of insufficient training samples.

A MULTIPLE FACE RECOGNITION SYSTEM WITH DLIB’S RESNET NETWORK USING DEEP METRIC LEARNING

Despite many approaches in facial recognition models, we often come across single facial recognition systems from an image.But the detection and recognition of a single face from an image is not very practical in an ever-changing world.For this, we need systems capable of detecting and recognizing multiple faces from a single image with which we can solve many real-world problems with a fewer number of images.So, we proposed an enhanced and efficient model, called HPMR (High Performance Multiple-face Recognition), to detect and recognize multiple faces from a single image. In this paper, we used Dlib’s ResNet network with 29 convolution layers to recognize faces. The network supports both Predictor 5 and Predictor 68 model to estimate facial landmarks. This model is trained on a data set of three million faces from well-known data sets like face scrub and VGG using deep metric learning. It uses HOG feature descriptor for multiple face detection. We used this model because it gives us better results compared to other existing models. Among many possible applications for multiple facial recognition, we have implemented an automated attendance system to demonstrate our approach.

Periocular Biometrics using Handcrafted Features for Non-Ideal Images

Features from face and iris to authenticate individuals are the most popular biometric traits. Still inclusion of non-ideal images (such as images with variation in pose, tilting head, subjects wearing spectacles and variation in capturing device distance) can degrade the recognition accuracy of any biometric systems. For this scenario, periocular region (nearby region around the eye) based biometric authentication is an emerging method which is used by researchers now a days to improve the recognition accuracy specifically for non-ideal images and when users are non- cooperative. In this context, our key insight is to develop a system considering periocular region as a biometric trait and aim to evaluate its effectiveness for classification of non-ideal images in two different non-ideal scenarios 1) images with different pose variation and 2) images captured from varying camera standoff distance. In this proposed work we have evaluated three different handcrafted feature descriptors 1) Histogram of Oriented Gradients 2) Bag of Feature model and 3)Local Binary Patterns on two different databases 1) ORL face database and 2) UBIPr periocular image database and found that HOG feature descriptor show superior performance as compare to BOF and LBP feature descriptor for periocular region based biometric authentication systems.

Optimal hog cell to image ratio for robust multi-sensor face recognition systems

The main problem for modern visible light face recognition has been accurate identification under variable environmental conditions. Thermal infrared facial images utilization in face recognition systems can provide a solution for problems related to uncontrolled environmental conditions, especially to those caused by illumination limitations. This paper compares the results of the use of visible light and thermal infrared imagery for face recognition based on the HOG feature descriptor. In particular, the paper suggests an optimal HOG cell to image size ratio in order to improve recognition accuracy and reduce computational complexity. Performance statistics are presented on facial images with different facial expressions. The obtained results support the conclusion that recognition with thermal infrared images is more robust and that fusion of sensors should be included for improving recognition accuracy.

Unconstrained face recognition in surveillance videos using moment invariants

Video-based face recognition is a challenging research area. In the unconstrained surveillance videos, recognition of faces with arbitrary poses, illumination, expressions, noise and blurriness is highly difficult task. In this paper, an efficient moment invariant face descriptor (MFD) is proposed. The geometric feature points of eyes, nose and mouth are extracted using face block filtering. These descriptors consist of 33 moment invariant features (11 from each). In these 11 moment invariant features, three of them are geometric moments and the remaining eight are central moments. It provides robustness against variability due to changes in localised regions of the faces. A public unconstrained video data set with three cameras is used. The proposed algorithm is compared with LBP, LPQ, HOG and moment invariant feature descriptors against different classifiers such as J48, random forest and multilayer perceptron. Experimental results show that the proposed algorithm using random forest classifier improves face recognition rate.