Search

COVID-19 originated in Wuhan, China, in December 2019 and quickly became a global outbreak in January 2020. COVID-19 is a disease caused by SARS-CoV-2, which is a human transmission disease. Since it is a human transmission disease, thus mass gathering in public is not allowed to prevent the possible spread of COVID-19. However, the current monitoring technology, such as closed-circuit television (CCTV), only cover a limited area of the public and lack of mobility. Image classification is one of the approaches that can detect crowds in an image and can be done through either machine learning or deep learning approach. Recently, deep learning, especially convolutional neural networks (CNNs) outperform classical machine learning in image classification and the common approach for modelling CNN is through transfer learning. Thus, this study aims to develop a convolutional neural network that can detect illegal crowd gathering from offline drone view images through image classification using the transfer learning technique. Several models are used to train on the same dataset obtained, and the all-model performance is evaluated through a confusion matrix. Based on performance analysis, it shows that the ResNet50 model outperforms the VGG16 model and InceptionV3 model by achieving 95% test accuracy, 95% precision, 95% recall and 95% F1-score. In conclusion, it can be concluded that the deep learning approach uses a pre-trained convolutional neural network that can be used to classify object images in this study.

Currency recognition has been widely developed using various types of techniques and able to assist people who have a visual impairment. Machine learning is one of the methods implemented where deep learning architecture is one of them. The deep learning approach is reliable and can be used in detection and recognition of objects based on images. As currency recognition has been developed for other currencies, thus in this project, currency recognition using Malaysian coins has been developed by modeling Convolutional Neural Network (CNN) in recognizing coin images. Malaysian coins dataset was developed consist of 2400 images of four classes of coins, 5 sen, 10 sen, 20 sen, and 50 sen. In this study, pretrained CNN which are AlexNet, GoogleNet, and MobileNetV2 were formulated in recognizing such coins. Performance of each trained model was evaluated using confusion matrix and GoogleNet obtained the best performance with 99.2% testing accuracy, 99.2% precision, 99.18% recall, and 99.19% F1 score. From the trained model, it can be further developed and implemented in assisting visually impaired persons by producing a prototype using Raspberry Pi and FPGA before it can be clinically tested on the subject.

Retinal tissue plays a crucial part in human vision. Infections of retinal tissue and delayed treatment or untreated infection could lead to loss of vision. Additionally, the diagnosis is prone to errors when huge dataset is involved. Therefore, a fully automated model of identification of retinal disease is proposed to reduce human interaction while retaining its high accuracy classification results. This paper introduces an enhanced design of a fully automatic multi-class retina diseases prediction system to assist ophthalmologists in making speedy and accurate investigation. Retinal fundus images, which have been used in this study, were downloaded from the stare website (157 images from five classes: BDR, CRVO, CNV, PDR, and Normal). The five files were categorized according to their annotations conducted by the experienced specialists. The categorized images were first processed with the proposed upgraded contrast-limited adaptive histogram filter for image brightness enhancement, noise reduction, and intensity spectrum normalization. The proposed model was designed with transfer learning method and the fine-tuned pre-trained RESNET50. Eventually, the proposed framework was examined with performance evaluation parameters, recorded a classification rate with 100% sensitivity, 100% specificity, and 100% accuracy. The performance of the proposed model showed a magnificent superiority as compared to the state-of-the-art studies.

The coronavirus disease 2019 (COVID-19) is a highly infectious disease caused by the SARS-CoV-2 coronavirus. In breaking the transmission chain of SARS-CoV-2, the government has made it compulsory for the people to wear a mask in public places to prevent COVID-19 transmission. Hence, an automated face mask detection is crucial to facilitate the monitoring process in ensuring people to wear a face mask in public. This project aims to develop an automated face and face mask detection for multiple people by applying deep learning-based object detection algorithm you only look once version 3 (YOLOv3). YOLOv3 object detection algorithm was concatenated with different backbones including ResNet-50 and Darknet-53 to develop the face and face mask detection model. Datasets were collected from online resources including Kaggle and Github and the images were filtered and labelled accordingly. The models were trained on 4393 images and evaluated based on precision, recall, mean average precision and the detection time. In conclusion, DarkNet53_YOLOv3 was chosen as the better model compared to ResNet50_YOLOv3 model with its good performance on accuracy with a mAP of 95.94% and a fast detection speed with a detection time of 50 seconds on 776 images.

Activity recognition in sports is often complex task resulting from the rapid dynamic interaction within players. In this paper, pre-trained VGG-16, deep learning based hockey activity recognition model has been proposed. Own hockey dataset consisting of four main activity includes free hit, goal, penalty corner and long corner was constructed as there are no existing field hockey datasets available. Experimental results indicate that the pre-trained deep learning model generates comparative results on this challenging dataset by tweaking the hyperparameters of this pre-trained model.

This paper proposes a stair walking detection via Long-short Term Memory (LSTM) network to prevent stair fall event happen by alerting caregiver for assistance as soon as possible. The tri-axial accelerometer and gyroscope data of five activities of daily living (ADLs) including stair walking is collected from 20 subjects with wearable inertial sensors on the left heel, right heel, chest, left wrist and right wrist. Several parameters which are window size, sensor deployment, number of hidden cell unit and LSTM architecture were varied in finding an optimized LSTM model for stair walking detection. As the result, the best model in detecting stair walking event that achieve 95.6% testing accuracy is double layered LSTM with 250 hidden cell units that is fed with data from all sensor locations with window size of 2 seconds. The result also shows that with similar detection model but fed with single sensor data, the model can achieve very good performance which is above 83.2%. It should be possible, therefore, to integrate the proposed detection model for fall prevention especially among patients or elderly in helping to alert the caregiver when stair walking event occur.

Sport is a competitive field, where it is an element of measurement for a countries development. Due to this reason, sport analysis has become one of the major contribution in analysing and improving the performance level of an athlete. Video-based modality has become a crucial tool used in sport analysis by coaches and performance analysis. There were wide variety of techniques used in sport video analysis. The main purpose of this review paper is to compare and update review between traditional handcrafted approach and deep learning approach in sport video analysis based on human activity recognition, overview of recent study in video based human activity recognition in sport analysis and finally concluded with future potential direction in sport video analysis.

The automatic retinal disease diagnosis by artificial intelligent is an interesting and challenging topic in the medical field. It requires an appropriate image enhancement technique and a sufficient training dataset for the specific retina conditions. The aim of this study was to design an automatic diagnosis convolutional neural network (CNN) model which does not require a large training dataset to specifically identify diabetic retinopathy symptoms, which are cotton wool, exudates spots and red lesionin colour fundus pictures. A novel framework comprised image enhancement method by using upgraded contrast limited adaptive histogram equalization (UCLAHE) filter and transferred pre-trained networks was developed to classify the retinal diseases regarding to the symptoms. The performance of the proposed framework was evaluated based on accuracy, sensitivity and specificity metrics. The collected results have proven the robustness of the proposed framework in offering good accuracy in retina diseases diagnosis.

Magnetic resonance imaging is an important modality in the diagnosis and pathology detection. Edge detection is used for image segmentation and feature extraction as part of the medical image analysis. There is no ideal and universal algorithm which performs perfectly under all conditions. Conventional Canny edge detector is not suitable to be used in Magnetic resonance images that contaminated by Rician noise. In this paper, we propose the use of customized non-local means into the Canny edge detector instead of Gaussian smoothing in the conventional Canny edge detector to effectively remove Rician noise while preserving edges in Magnetic resonance image of an internal organ. The result shows that our method can yield better edge detection than conventional method, with minimal false edge detection. The proposed method undergoes several attempts of parameter adjustment to detect true edges successfully using optimal parameter setting.

Sport activity recognition has become one of the primary sport performance analysis contributions as it offers a notational data for tactical analysis and planning. In developing of activity recognition algorithm, searching for the best window segmentation size within time-series data is one of the parameters that contribute to the performance of algorithm in term of the accuracy rate. Yet, previous studies on activity recognition have implemented a different fixed size of window in segmenting the specific activities and the performance of activity recognition algorithm along with the changes of window size is still uncertain. Thus, this study is investigating the performance of field hockey activity recognition algorithm based on inertial sensor time-series data with different window segmentation size. The study was conducted on 11 subjects who worn the inertial sensors on their chest and waist while performing the six common field hockey activities which are passing, drive, drag flick, dribbling, receiving and tackles. The performance of each size of windows were observed and evaluated by using Cubic support vector machine (SVM). Among the different window sizes studied, this study found 1.5 s and 2.0 s are among the top in producing high accuracy rate for recognizing the field hockey activity that represent the movement from chest and waist with 89.6% and 91.4% accuracy respectively.

Performance analysis is essential in sports practice where the athlete is evaluated to improve their performance. Due to the rapid growth of science and technology, research on automated recognition of sports actions has become ubiquitous. The implementation of automated action recognition is an effort to overcome the manual action recognition in sport performance analysis. In this study, we developed a model for automated badminton action recognition from the computer vision data inputs using the deep learning pre-trained AlexNet Convolutional Neural Network (CNN) for features extraction and classify the features using supervised machine learning method which is linear Support-Vector Machine (SVM). The data inputs consist of badminton match images of two classes: hit and non-hit action. Before pre-trained AlexNet CNN was directly extracting the features, we introduced the new local CNN extractor in recognition pipeline. The results show that the classification accuracy with this new local CNN method achieved 98.7%. In conclusion, this new local CNN extractor can contribute to the improvement of the performance accuracy of the classification task.

Deep learning approach has becoming a research interest in action recognition application due to its ability to surpass the performance of conventional machine learning approaches. Convolutional Neural Network (CNN) is among the widely used architecture in most action recognition works. There are various models exist in CNN but no research has been done to analyse which model has the best performance in recognizing actions for badminton sport. Hence, in this paper we are comparing the performance of four different pre-trained models of deep CNN in classifying the badminton match images to recognize the different actions done by the athlete. Four models used for comparison are AlexNet, GoogleNet, VggNet-16 and VggNet-19. The images used in this experimental work are categorized into two classes: hit and non-hit action. Firstly, each image frame was extracted from Yonex All England Man Single Match 2017 broadcast video. Then, the image frames were fed as the input to each classifier model for classification. Finally, the performance of each classifier model was evaluated by plotting its performance accuracy in form of confusion matrix. The result shows that the GoogleNet model has the highest classification accuracy which is 87.5% compared to other models. In a conclusion, the pre-trained GoogleNet model is capable to be used in recognizing actions in badminton match which might be useful in badminton sport performance technology.

Visually impaired people faced a problem in identifying and recognizing the different types of banknote due to some reasons. This problem draws researchers’ attention to introduce an automated banknote recognition system that can be divided into a vision-based system and sensor-based system. The main aim of this study is to have deeper analysis on the effect of region and orientation on the performance of Machine Learning and Deep Learning respectively using Malaysian Ringgit banknotes (RM 1, RM 5, RM 10, RM 20, RM 50 and RM 100). In this project, two experiments conducted on two types of banknote image: different region and orientation captured by using handphone camera in a controlled environment. Feature extraction of the RGB values called RB, RG, and GB from banknote image with different region were used to the machine learning classification algorithms such as k-Nearest Neighbors (kNN) and Decision Tree Classifier (DTC), Support Vector Machine (SVM) and Bayesian Classifier (BC) for recognizing each class of banknote. Banknote image with different orientation was directly feed to AlexNet, a pre-trained model of Convolutional Neural Network (CNN), the most popular image processing structure of Deep Learning Neural Network. Ten-fold cross-validation was used to select the optimized kNN, DTC, SVM, and BC which was based on the smallest cross-validation loss. After that, the performance of kNN, DTC, SVM, BC and AlexNet model was presented in a confusion matrix. Both kNN and DTC achieved 99.7% accuracy but both SVM and BC perform better by succeeded to achieve 100% accuracy. It also can be concluded that AlexNet can only perform great in testing new data if only the data had previously been trained with similar orientation. Orientation does give effect to the performance of AlexNet model.

Sport performance analysis in sports practice cannot be separable. It is important to help coach analyse and improve the performance of their athletes through training or game session. Due to the advancement of technology nowadays, the notational analysis of the video content using various software packages has become possible. Unluckily, the coach needs to recognize the actions manually before doing further analysis. The purpose of this study is to formulate an automated system for badminton smash recognition on widely available broadcasted videos using pre-trained Convolutional Neural Network (CNN) method. Smash and other badminton actions such as clear, drop, lift and net from the video were used to formulate the CNN models. Therefore, two experiments were conducted in this study. The first experiment is the study on the performance between four different existing pre-trained models which is AlexNet, GoogleNet, Vgg-16 Net and Vgg-19 Net in recognizing five actions. The results show that the pre-trained AlexNet model has the highest performance accuracy and fastest training period among the other models. The second experiment is the study on the performance of two different pre-trained models which is AlexNet and GoogleNet to recognize smash and non-smash action only. The results show that the pre-trained GoogleNet model produces the best performance in recognizing smash action. In conclusion, pre-trained AlexNet model is suitable to be used to automatically recognize the five badminton actions while GoogleNet model is excellent at recognizing smash action from the broadcasted video for further notational analysis.

Having a systemic system in recognizing activity in sports is very essential along with enhancing the performance analysis in sport. As the system is required to provide a quality, reliable and unbiased notational data for determining the strength and weakness of field hockey players. Therefore, this study is analysing the accelerometer and gyroscope signal on of the four inertial sensors attached to the upper body chest, waist, right and left wrist and formulate the best model in using the wearable sensor for human activity recognition in the field hockey which are passing, drive, drag flick, dribbling, receiving and tackling. Set of features such as mean, standard deviation, maximum and minimum peak are extracted from each inertial sensor signal as an input vector for classification purpose. Results from the study shows that the recognition using combination of all four sensors achieved the highest performance of 96.7% accuracy; and waist and left wrist is recommended if single sensor based human activity recognition is preferred.

This paper aims to identify common sport activity recognition by using inertial sensor. Various classifiers from the Classification Learner App in MATLAB is used to recognize the common sport activity. In data collection, 10 subjects are asked to perform several common sport activities which is stationary, walking, jogging, sprinting, and jumping. The data is collected by using Physilog® 4 Silver which is an inertial sensor from Gait Up that is attached to the chest of the subject using strap. 3D accelerometer and gyroscope data from inertial sensor is observed and processed in this study. To recognize the common sport activity from the data, time-domain features such as mean, standard deviation, maximum and minimum value is extracted from the signal produced by inertial sensor. After that, the data is trained using the selected features and 10-fold cross validation is applied to set the training and testing data. The confusion matrix and accuracy result from each classifier is observed. In conclusion, Support Vector Machine with Cubic SVM is selected as the best model in recognizing common sport activity as it produced the highest accuracy, 91.2% in the confusion matrix. This classifier performance outperforms other classifiers in detecting five types of common sport activity from the data collected.

3D shape descriptor has been used widely in the field of 3D object retrieval. However, the performance of object retrieval greatly depends on the shape descriptor used. The aims of this study is to review and compare the common 3D shape descriptors proposed in 3D object retrieval literature for object recognition and classification based on Kinect-like depth image obtained from RGB-D object dataset. In this paper, we introduce (1) inter-class; and (2) intra-class evaluation in order to study the feasibility of such descriptors in object recognition. Based on these evaluations, local spin image outperforms the rest in discriminating different classes when several depth images from an instance per class are used in inter-class evaluation. This might be due to the slightly consistent local shape property of such images and due to the proposed local similarity measurement that manages to extract the local based descriptor. However, shape distribution performs excellent for intra-class evaluation (that involves several instances per class) may be due to the global shape from different instances per class is slightly unchanged. These results indicate a remarkable feasibility analysis of the 3D shape descriptor in object recognition that can be potentially used for Kinect-like sensor.

Nowadays, coaches and sport analyst are concerning about sport performance analysis through sport video match. However, they still used conventional method which is through manual observation of the full video that is very troublesome because they might miss some meaningful information presence in the video. Several previous studies have discussed about tracking ball movements, identification of player based on jersey color and number as well as player movement detection in various type of sport such as soccer and volleyball but not in badminton. Therefore, this study focused on developing an automated system using Faster Region Convolutional Neural Network (Faster R-CNN) to track the position of the badminton player from the sport broadcast video. In preparing the dataset for training and testing, several broadcast videos were converted into image frames before labelling the region which indicate the players. After that, several different trained Faster R-CNN detectors were produced from the dataset before tested with different set of videos to evaluate the detector performance. In evaluating the performance of each detector model, the average precision was obtained from precision recall graph. As a result, this study revealed that the detector successfully detects the player when the detector is being fed with more generalized dataset.

Sport performance analysis which is crucial in sport practice is used to improve the performance of athletes during the games. Many studies and investigation have been done in detecting different movements of player for notational analysis using either sensor based or video based modality. Recently, vision based modality has become the research interest due to the vast development of video transmission online. There are tremendous experimental studies have been done using vision based modality in sport but only a few review study has been done previously. Hence, we provide a review study on the video based technique to recognize sport action toward establishing the automated notational analysis system. The paper will be organized into four parts. Firstly, we provide an overview of the current existing technologies of the video based sports intelligence systems. Secondly, we review the framework of action recognition in all fields before we further discuss the implementation of deep learning in vision based modality for sport actions. Finally, the paper summarizes the further trend and research direction in action recognition for sports using video approach. We believed that this review study would be very beneficial in providing a complete overview on video based action recognition in sports.

Shape distribution is a common 3D shape descriptor that has been widely used for 3D object retrieval. In this study, we evaluate the feasibility of shape distribution for object recognition based on Kinect-like depth image obtained from RGB-D object dataset; consisting of several household instances from 51 classes; and each instance consists of depth images from different rotational angle view. The proposed evaluation procedures are called (1) inter-class evaluation and (2) intra-class evaluation and were used to evaluate the shape distribution performance. Based on these evaluations, shape distribution performance was found to be slightly decreased in inter-class manner while significantly decreased for intra-class. It is evident that the minimal performance degradation in inter-class evaluation is due to variety of formed shapes when the instance is rotated while shape distribution suffers from not only shape variation among different rotational angle view but also among different instance per class in intra-class evaluation. This preliminary attempt shows that shape distribution is a relevant candidate applicable in object recognition for Kinect-like depth image.

The rapid development of intelligent assistive technology for replacing a human caregiver in assisting people with dementia performing activities of daily living (ADLs) promises in the reduction of care cost especially in training and hiring human caregiver. The main problem however, is the various kinds of sensing agents used in such system and is dependent on the intent (types of ADLs) and environment where the activity is performed. In this paper on overview of the potential of computer vision based sensing agent in assistive system and how it can be generalized and be invariant to various kind of ADLs and environment. We find that there exists a gap from the existing vision based human action recognition method in designing such system due to cognitive and physical impairment of people with dementia.

In this paper, we propose a new hybrid approach for image segmentation. The proposed approach exploits spatial fuzzy c-means for clustering image pixels into homogeneous regions. In order to improve the performance of fuzzy c-means to cope with segmentation problems, we employ gravitational search algorithm which is inspired by Newton’s rule of gravity. Gravitational search algorithm is incorporated into fuzzy c-means to take advantage of its ability to find optimum cluster centers which minimizes the fitness function of fuzzy c-means. Experimental results show effectiveness of the proposed method in segmentation different types of images as compared to classical fuzzy c-means.

We introduce in this paper, a shadow detection and removal method for moving objects especially for humans and vehicles. An effective method is presented for detecting and removing shadows from foreground figures. We assume that the foreground figures have been extracted from the input image by some background subtraction method. A figure may contain only one moving object with or without shadow. The homogeneity property of shadows is explored in a novel way for shadow detection and image division technique is used. The process is followed by filtering, removal, boundary removal and removal validation.