Aggregate Channel Features Research Articles

Fast Rock Detection in Visually Contaminated Mining Environments Using Machine Learning and Deep Learning Techniques

Advances in machine learning algorithms have allowed object detection and classification to become booming areas. The detection of objects, such as rocks, in mining operations is affected by fog, snow, suspended particles, and high lighting. These environmental conditions can stop the development of mining work, which entails a considerable increase in operating costs. It is vital to select a machine learning algorithm that is accurate, fast, and contributes to lower operational costs because of the aforementioned environmental situations. In this study, the Viola-Jones algorithm, Aggregate Channel Features (ACF), Faster Regions with Convolutional Neural Networks (Faster R-CNN), Single-Shot Detector (SSD), and You Only Look Once (YOLO) version 4 were analyzed, considering the precision metrics, recall, AP50, and average detection time. In our preliminary tests, we have observed that the differences between YOLO v4 and the latest versions are not substantial for the specific problem of rock detection addressed in our article. Therefore, YOLO v4 is an appropriate and representative choice for evaluating the effectiveness of existing methods in our study. The YOLO v4 algorithm performed the best overall, whereas the SSD algorithm performed the fastest. The results indicate that the YOLO v4 algorithm is a promising candidate for detecting rocks with visual contamination in mining operations.

Development of a Hybrid Method for Multi-Stage End-to-End Recognition of Grocery Products in Shelf Images

Product recognition on grocery shelf images is a compelling task of object detection because of the similarity between products, the presence of the different scale of product sizes, and the high number of classes, in addition to constantly renewed packaging and added new products’ difficulty in data collection. The use of conventional methods alone is not enough to solve a number of retail problems such as planogram compliance, stock tracking on shelves, and customer support. The purpose of this study is to achieve significant results using the suggested multi-stage end-to-end process, including product detection, product classification, and refinement. The comparison of different methods is provided by a traditional computer vision approach, Aggregate Channel Features (ACF) and Single-Shot Detectors (SSD) are used in the product detection stage, and Speed-up Robust Features (SURF), Binary Robust Invariant Scalable Key points (BRISK), Oriented Features from Accelerated Segment Test (FAST), Rotated Binary Robust Independent Elementary Features (BRIEF) (ORB), and hybrids of these methods are used in the product classification stage. The experimental results used the entire Grocery Products dataset and its different subsets with a different number of products and images. The best performance was achieved with the use of SSD in the product detection stage and the hybrid use of SURF, BRISK, and ORB in the product classification stage, respectively. Additionally, the proposed approach performed comparably or better than existing models.

Aggregate Channel Features and Fast Regions CNN Approach for Classification of Ship and Iceberg

Detection and classification of icebergs and ships in synthetic aperture radar (SAR) images play a vital role in marine surveillance systems even though available adaptive threshold methods give satisfying results on detection and classification for ships and icebergs, including techniques of convolutional neural networks (CNNs), but need more accuracy and precision. An efficient and accurate method was developed to detect and classify the ship and icebergs. Hence, the research method proposed locating and classifying both ships and icebergs in a given SAR image with the help of deep learning (DL) and non-DL methods. A non-DL method utilized here was the aggregate channel features (ACF) detector, which extracts region proposals from huge SAR images. The DL object detector called fast regions CNN (FRCNN) detects objects accurately from the result of ACF since the ACF method avoids unwanted regions. The novelty of this study was that ACF-FRCNN concentrates only on accurately classifying ships and icebergs. The proposed ACF-FRCNN method gave a better performance in terms of loss (18.32%), accuracy (96.34%), recall (98.32%), precision (95.97%), and the F1 score (97.13%). Compared to other conventional methods, the combined effect of ACF and FRCNN increased the speed and quality of the detection of ships and icebergs. Thus, the ACF-FRCNN method is considered a novel method for over 75 × 75 resolution ship and iceberg SAR images.

Multi – oil droplet recognition of oil-water two-phase flow based on integrated features

Multi-oil droplet target recognition is one of the applications of machine vision in the measurement of oil-water two-phase flow parameters, which could combine other algorithms to obtain the oil droplet velocity and the water holdup of oil water two-phase flow. Appropriate target representation features can improve the recognition effect of multiple oil droplets. However, due to shooting environment differences and quality differences of oil-water two-phase flow images, existing target representation features do not perform well in low-quality oil-water two-phase flow images. To improve the precision of multi-oil droplet target recognition in oil-water two-phase flow and reduce the miss rate, this paper constructs an integrated feature on the basis of aggregate channel features (ACF). The integrated feature named aggregate channel features with histogram of local gravitational feature（ACFHG） contains the color feature channels reflecting the overall color features of the oil droplet sample, the gradient amplitude channel reflecting the overall gradient of the oil droplet sample image, the gradient direction histogram feature channels reflecting the local gradient of the oil droplet sample image, and the local gravitational feature channels that ensure oil droplet target recognition in low quality photos and photos taken in complex shooting environments. Moreover, the rotation invariance is obtained by taking the oriented gradient histogram of the local gravitational feature to further improve the multi-oil droplet target recognition effect. Experiment results show that the average precision of multi-oil droplet target recognition using the integrated features is 83.38%, which is 9.93% higher than that with using ACF, and the miss rate is 9.13%, which is 57.18% lower than that with using ACF. Compared with other existing target detection methods, the method proposed in this paper still has an advantage in the rate of missed detection.

Investigation and Analysis of Crack Detection using UAV and CNN: A Case Study of Hospital Raja Permaisuri Bainun

Crack detection in old buildings has been shown to be inefficient, with many technical challenges such as physical inspection and difficult measurements. It is important to have an automatic, fast visual inspection of these building components to detect cracks by evaluating their conditions (impact) and the level of their risk. Unmanned Aerial Vehicles (UAV) can automate, avoid visual inspection, and avoid other physical check-ups of these buildings. Automated crack detection using Machine Learning Algorithms (MLA), especially a Conventional Neural Network (CNN), along with an Unmanned Aerial Vehicle (UAV), can be effective and both can efficiently work together to detect the cracks in buildings using image processing techniques. The purpose of this research project is to evaluate currently available crack detection systems and to develop an automated crack detection system using Aggregate Channel Features (ACF)that can be used with unmanned aerial vehicles (UAV). Therefore, we conducted a real-world experiment of crack detection at Hospital Raja Permaisuri Bainun using DJI Mavic Air (Drone Hardware) and DJI GO 4(Drone Software) using CNN through MATLAB software with CNN-SVM method with the accuracy rate of3.0 percent increased from 82.94%to 85.94%. in comparison with other ML algorithms like CNN Random Forest (RF), Support Vector Machine (SVM) and Artificial Neural Network (ANN).

Fully automated method for dental age estimation using the ACF detector and deep learning

BackgroundDental age estimation plays an important role in identifying an unknown person. In forensic science, estimating age with high accuracy depends on the experience of the practitioner. Previous studies proposed classification of tooth development of the mandibular third molar by following Demirjian’s method, which is useful for dental age estimation. Although stage of tooth growth is very helpful in assessing age estimation, it must be performed manually. The drawback of this procedure is its need for skilled observers to carry out the tasks precisely and reproducibly because it is quite detailed. Therefore, this research aimed to apply computer-aid methods for reducing time and subjectivity in dental age estimation by using dental panoramic images based on Demirjian’s method. Dental panoramic images were collected from persons aged 15 to 23 years old. In accordance with Demirjian’s method, this study focused only on stages D to H of tooth development, which were discovered in the 15- to 23-year age range. The aggregate channel features detector was applied automatically to localize and crop only the lower left mandibular third molar in panoramic images. Then, the convolutional neural network model was applied to classify cropped images into D to H stages. Finally, the classified stages were used to estimate dental age.ResultsExperimental results showed that the proposed method in this study can localize the lower left mandibular third molar automatically with 99.5% accuracy, and training in the convolutional neural network model can achieve 83.25% classification accuracy using the transfer learning strategy with the Resnet50 network.ConclusionIn this work, the aggregate channel features detector and convolutional neural network model were applied to localize a specific tooth in a panoramic image and identify the developmental stages automatically in order to estimate the age of the subjects. The proposed method can be applied in clinical practice as a tool that helps clinicians to reduce the time and subjectivity for dental age estimation.

Aggregate Channel Features for newborn face detection in Neonatal Intensive Care Units.

An efficient face detector could be very helpful to point out possible neurological dysfunctions such as seizure events in Neonatal Intensive Care Units. However, its development is still challenging because large public datasets of newborns' faces are missing. Over the years several studies introduced semi-automatic approaches. This study proposes a fully automated face detector for newborns in Neonatal Intensive Care Units, based on the Aggregate Channel Feature algorithm. The developed method is tested on a dataset of video recordings from 42 full-term newborns collected at the Neuro-physiopathology and Neonatology Clinical Units, AOU Careggi, Firenze, Italy. The proposed system showed promising results, giving (mean ± standard error): log-Average Miss Rate = 0.47 ± 0.05 and Average Precision Recall = 0.61 ± 0.05. Moreover, achieved results highlighted interesting differences between newborns without seizures, newborns with electro-clinical seizures, and newborns with electrographic-only seizures. For both metrics statistically significant differences were found between patients with electro-clinical seizures and the other two groups. Clinical Relevance- The proposed method, based on quantitative physio-pathological features of facial movements, is of clinical relevance as it could speed up pain or seizure assessment of newborns in Neonatal Intensive Care Units.

Fully automated determination of the cervical vertebrae maturation stages using deep learning with directional filters.

IntroductionWe aim to apply deep learning to achieve fully automated detection and classification of the Cervical Vertebrae Maturation (CVM) stages. We propose an innovative custom-designed deep Convolutional Neural Network (CNN) with a built-in set of novel directional filters that highlight the edges of the Cervical Vertebrae in X-ray images.MethodsA total of 1018 Cephalometric radiographs were labeled and classified according to the Cervical Vertebrae Maturation (CVM) stages. The images were cropped to extract the cervical vertebrae using an Aggregate Channel Features (ACF) object detector. The resulting images were used to train four different Deep Learning (DL) models: our proposed CNN, MobileNetV2, ResNet101, and Xception, together with a set of tunable directional edge enhancers. When using MobileNetV2, ResNet101 and Xception, data augmentation is adopted to allow adequate network complexity while avoiding overfitting. The performance of our CNN model was compared with that of MobileNetV2, ResNet101 and Xception with and without the use of directional filters. For validation and performance assessment, k-fold cross-validation, ROC curves, and p-values were used.ResultsThe proposed innovative model that uses a CNN preceded with a layer of tunable directional filters achieved a validation accuracy of 84.63%84.63% in CVM stage classification into five classes, exceeding the accuracy achieved with the other DL models investigated. MobileNetV2, ResNet101 and Xception used with directional filters attained accuracies of 78.54%, 74.10%, and 80.86%, respectively. The custom-designed CNN method also achieves 75.11% in six-class CVM stage classification. The effectiveness of the directional filters is reflected in the improved performance attained in the results. If the custom-designed CNN is used without the directional filters, the test accuracy decreases to 80.75%. In the Xception model without the directional filters, the testing accuracy drops slightly to 79.42% in the five-class CVM stage classification.ConclusionThe proposed model of a custom-designed CNN together with the tunable Directional Filters (CNNDF) is observed to provide higher accuracy than the commonly used pre-trained network models that we investigated in the fully automated determination of the CVM stages.

Static Hand Gesture Angle Recognition via Aggregated Channel Features (ACF) Detector

Static hand gesture recognition is critical in the development of a system for human-computer interaction. Many human-computer interactions, such as human-robot interaction, game control, control of smart home devices, and others, use hand gestures as a fundamental and natural language of the body. The direction of rotation of static hand gestures is the subject of this research, and the focus is on six degrees of rotation (0°, 45°, 90°, 180°, 270°, and 315°). This work presents an ideal approach that can recognize the angle of hand gestures based on the Aggregate Channel Features (ACF) detector. This approach consists of three main stages: preprocessing (image labelling), computer training, and hand angle detection based on the ACF detector. The training process consists of 25 stages. The static hand gesture dataset contained 569 images (361 for training and 208 for testing). The average time cost to detect all hand gesture angles was 0.9445 seconds, and all hand angles were recognized with 100% accuracy. This is a strong indication that supports our approach.

Pedestrian Detection Using Integrated Aggregate Channel Features and Multitask Cascaded Convolutional Neural-Network-Based Face Detectors.

Pedestrian detection is a challenging task, mainly owing to the numerous appearances of human bodies. Modern detectors extract representative features via the deep neural network; however, they usually require a large training set and high-performance GPUs. For these cases, we propose a novel human detection approach that integrates a pretrained face detector based on multitask cascaded convolutional neural networks and a traditional pedestrian detector based on aggregate channel features via a score combination module. The proposed detector is a promising approach that can be used to handle pedestrian detection with limited datasets and computational resources. The proposed detector is investigated comprehensively in terms of parameter choices to optimize its performance. The robustness of the proposed detector in terms of the training set, test set, and threshold is observed via tests and cross dataset validations on various pedestrian datasets, including the INRIA, part of the ETHZ, and the Caltech and Citypersons datasets. Experiments have proved that this integrated detector yields a significant increase in recall and a decrease in the log average miss rate compared with sole use of the traditional pedestrian detector. At the same time, the proposed method achieves a comparable performance to FRCNN on the INRIA test set compared with sole use of the Aggregated Channel Features detector.

Automatic modelling of urban subsurface with ground-penetrating radar using multi-agent classification method

ABSTRACT The subsurface of urban cities is becoming increasingly congested. In-time records of subsurface structures are of vital importance for the maintenance and management of urban infrastructure beneath or above the ground. Ground-penetrating radar (GPR) is a nondestructive testing method that can survey and image the subsurface without excavation. However, the interpretation of GPR relies on the operator’s experience. An automatic workflow was proposed for recognizing and classifying subsurface structures with GPR using computer vision and machine learning techniques. The workflow comprises three stages: first, full-cover GPR measurements are processed to form the C-scans; second, the abnormal areas are extracted from the full-cover C-scans with coefficient of variation-active contour model (CV-ACM); finally, the extracted segments are recognized and classified from the corresponding B-scans with aggregate channel feature (ACF) to produce a semantic map. The selected computer vision methods were validated by a controlled test in the laboratory, and the entire workflow was evaluated with a real, on-site case study. The results of the controlled and on-site case were both promising. This study establishes the necessity of a full-cover 3D GPR survey, illustrating the feasibility of integrating advanced computer vision techniques to analyze a large amount of 3D GPR survey data, and paves the way for automating subsurface modeling with GPR.

Pedestrian Behavior Analytics on Dashcam Videos in Chaotic Environments

Pedestrian safety is a major challenge in geographies where the traffic is not well managed, such as developing nations. Pedestrians often lose their lives when moving vehicles or buses hit them, for example, nearly half the road mishaps in Delhi involve such scenarios. We envision that video analytics on dashboard cameras can solve this problem if the pedestrians could be detected and tracked within permissible limits to alert the vehicle drivers. In this paper, we propose a novel end-to-end pipeline for multi-pedestrian detection and tracking their trajectories by leveraging multi-scale aggregate channel features in a tracking-by-detection paradigm and temporal trajectory analysis for pedestrian trajectory characterization to identify if the pedestrian is on a potential collision course with the vehicle. We evaluate this framework on a novel dataset collected from dashboard cameras with frontal view of Delhi roads with unlaned and chaotic traffic conditions. The problem of multi-pedestrian detection, tracking and characterizing pedestrian behavior on the road becomes challenging because of the motion of camera mounted on-board the vehicle and different traffic scenarios on Delhi roads. Extensive experiments demonstrate that the proposed method achieves promising results over several pedestrian detection baselines on the collected dataset with an average precision of 94.71% in detection and an accuracy of 86.48% in pedestrian behavior classification.

Effectiveness of Human Detection from Aerial Images Taken from Different Heights

Recently, drones have been regularly used to aid in search and rescue in places where it is normally to carry out some of the early forensic victim localization. There are many suitable human detectors for drone use, such as Histogram Oriented Gradient (HOG), You Only Looks Once (YOLO), and Aggregate Channel Features (ACF). In this paper, the height of the aerial images is analyzed for its effect on the accuracy of the detection. This works compares ACF, YOLO MobileNet, and YOLO ResNet50 using a different set of aerial images varying at 10m, 20m, and 30m heights. The results show that in a single-model test, with our proposed bounding-box standardization, YOLO MobileNet achieves significant increase in test precision (AP), with 0.7 AP recorded. For single-model test, YOLO MobileNet yield best AP using 20m training data where it obtained AP of 0.88 (10m test height), 0.82 (20m test height), and 0.91 (30m test height).

Object Visual Detection for Intelligent Vehicles

Traffic scene perception (TSP) aims to extract accurate real-time on-road environment information, which involves three phases detection of objects of interest, recognition of detected objects, and tracking of objects in motion. In this project, we focus on three important classes of objects: traffic signs, cars, and cyclists. We propose to detect all the three important objects in a single learning-based detection framework. To enhance the feature robustness to noises and image deformations, we introduce spatially pooled features as a part of aggregated channel features. In order to further improve the generalization performance, we propose an object sub categorization method as a means of capturing the intra class variation of objects. We experimentally demonstrate the effectiveness and efficiency of the proposed framework in three detection applications: traffic sign detection, car detection, and cyclist detection.

Visual-Based Motorcycle Detection using You Only Look Once (YOLO) Deep Network

In Malaysia and various Asian countries, use of motorcycle is growing in numbers, and the motorcycle is the dominating transport mode. In fact, the number of motorcycles per thousand people averaged over several major Asian cities is significantly higher than the average of the rest of the world. With growing use of motorcycle, blind spots have become one of major cause to road injuries and fatalities of motorcyclists in Malaysia. In this work, a visual-based detector is proposed to reduce the risk of blind spots in causing road injuries to motorcyclists. The objective of this detector is to provide alerts to the drivers of cars and other vehicles when there are motorcyclists in proximity to the vehicles especially around blind spot areas. In developing this solution, a variant of deep network called You Only Look Once (YOLO) is chosen as visual-based detector. This YOLO deep network is trained and tested using the 5811 collected images of motorcyclists. A benchmark with regards to its accuracy and speed is conducted by comparing this visual detector against several methods such as Aggregate Channel Features (ACF) and Faster Region Convolutional Neural Network (FRCNN). Results showed that YOLO detector is the most superior detector since it has the best average precision out of all detectors, and its inference time at 22.55 ms (44 fps) is able to provide real-time implementation. Besides, YOLO inference on machines without GPU still manage to achieve a commanding performance, which is on average, at 17 fps.

Pedestrian- and Vehicle-Detection Algorithm Based on Improved Aggregated Channel Features

In advanced driver-assistance systems (ADAS), the accuracy and real-time performance of pedestrian- and vehicle-detection algorithms based on vision sensors are crucial for safety. Here, a lightweight detection algorithm based on aggregated channel features (ACFs),consisting of a context pixel ACF (CP-ACF) pedestrian detector and a multiview ACF (Mv-ACF) vehicle detector, is proposed to rapidly and precisely understand road scenes. The former fuses local and context information to improve the robustness to pedestrian deformation, while the latter contains a number of subclass detectors to alleviate intraclass differences due to different viewing angles. Compared to the original ACF, the CP-ACF pedestrian detector reduces the average miss rate (AMR) by 6.34%. The Mv-ACF vehicle detector improves the average precision (AP) by 40.26% on average at easy, moderate and hard levels. This remarkable effectiveness is due to the spectrum clustering of multiview samples and the resulting integration of these subclass detectors via confidence score calibration, which reduces the intraclass differences of vehicles. Since feature extraction takes up 68.8% of the total detection time, a mechanism of feature sharing between pedestrian and vehicle detectors is advanced to reduce the time spent in feature extraction. A strategy based on ground-plane constraints (GPCs) is proposed to control false detection of pedestrians and vehicles by incorporating road prior information, which reduces the AMR by 1.07% for CP-ACF pedestrian detectors and improves the AP by 0.27% on average for Mv-ACF vehicle detectors. Thus, the proposed algorithm can effectively control false detection by road prior information.

Convolutional Neural Network Based Approach to Detect Pedestrians in Real-Time videos

Pedestrians in the vehicle way are in peril of being hit, along these lines making extreme damage walkers and vehicle inhabitants. Hence, constant person on foot identification was done through a set of recorded videos and the system detects the persons/pedestrians in the given input videos. In this survey, a continuous plan was proposed dependent on Aggregated Channel Features (ACF) and CPU. The proposed technique doesn’t have to resize the information picture neither the video quality. We also use SVM with HOG and SVM with HAAR to detect the pedestrians. In addition, the Convolutional Neural Networks (CNN) were trained with a set of pedestrian images datasets and later tested on some test-set of pedestrian images. The analyses demonstrated that the proposed technique could be utilized to distinguish people on foot in the video with satisfactory mistake rates and high prediction accuracy. In this manner, it tends to be applied progressively for any real-time streaming of videos and also for prediction of pedestrians in pre-recorded videos.

A Method for Identification of Face of the Person and Presence of Facial Spot in the Face

Human Face and facial parts are the most significant parts as it reveals a person’s true identity. It plays an important role in various biometric applications like crowd analysis, human tracking, photography, cosmetic surgery, etc. There are many techniques are available to detect a facial image. Among them, skin detection is the most popular one. The aim of this paper is to detect first the person's identity from facial image and finally check any spot present the detected person. The first step is to detect the maximum skin region based on a combination method of RGB and HSV color space model. Next it is to verify the skin areas of human through machine learning approach. The Aggregated Channel Features (ACF) detector is used to identify the different facial parts like eye pairs, nose, and mouth. Bootstrap aggregation decision tree classifier is applied to classify the person’s identity based on Histogram Oriented Gradient (HOG) features value. The experimental results show that the proposed method gives the average 97% accuracy.

Flue gas layer feature segmentation based on multi-channel pixel adaptive

It is very difficult to accurately separate the smoke contour in fire video. Because the scene of a fire is complex and changeable, and there are often many interference factors, such as continuously changing light and fast switching scenes, it is difficult to accurately guarantee the separation of smoke contours. In this paper, an aggregate channel feature algorithm that combines color, saturation and texture is designed, and a fast pixel block feature matching method is used to build the background model. In order to overcome the error caused by scene switching, a dynamic threshold control method based on the background switching speed is proposed, which eliminates the interference caused by the dynamic background update, and effectively extracts the foreground smoke pixels and smoke contour map. The experimental results show that the algorithm can accurately extract the smoke layer contour map, and compared with the traditional foreground extraction algorithms, the algorithm is faster and more accurate.

Tuning of Classifiers to Speed-Up Detection of Pedestrians in Infrared Images.

This paper presents an experimental evaluation of real-time pedestrian detection algorithms and their tuning using the proposed universal performance index. With this index, the precise choice of various parameters is possible. Moreover, we determined the best resolution of the analysis window, which is much lower than the initial window. By such means, we can speed-up the processing (i.e., reduce the classification time by 74%). There are cases in which we increased both the processing speed and the classification accuracy. We made experiments with various baseline detectors and datasets in order to confirm versatility of the proposed ideas. The analyzed classifiers are those typically applied to detection of pedestrians, namely: aggregated channel feature (ACF), deep convolutional neural network (CNN), and support vector machine (SVM). We used a suite of five precisely chosen night (and day) IR vision datasets.