RESUMEN

This review aims to critically examine the existing state-of-the-art forest fire detection systems that are based on deep learning methods. In general, forest fire incidences bring significant negative impact to the economy, environment, and society. One of the crucial mitigation actions that needs to be readied is an effective forest fire detection system that are able to automatically notify the relevant parties on the incidence of forest fire as early as possible. This review paper has examined in details 37 research articles that have implemented deep learning (DL) model for forest fire detection, which were published between January 2018 and 2023. In this paper, in depth analysis has been performed to identify the quantity and type of data that includes images and video datasets, as well as data augmentation methods and the deep model architecture. This paper is structured into five subsections, each of which focuses on a specific application of deep learning (DL) in the context of forest fire detection. These subsections include 1) classification, 2) detection, 3) detection and classification, 4) segmentation, and 5) segmentation and classification. To compare the model's performance, the methods were evaluated using comprehensive metrics like accuracy, mean average precision (mAP), F1-Score, mean pixel accuracy (MPA), etc. From the findings, of the usage of DL models for forest fire surveillance systems have yielded favourable outcomes, whereby the majority of studies managed to achieve accuracy rates that exceeds 90%. To further enhance the efficacy of these models, future research can explore the optimal fine-tuning of the hyper-parameters, integrate various satellite data, implement generative data augmentation techniques, and refine the DL model architecture. In conclusion, this paper highlights the potential of deep learning methods in enhancing forest fire detection that is crucial for forest fire management and mitigation.

RESUMEN

One of the most common and deadly diseases in the world is lung cancer. Only early identification of lung cancer can increase a patient's probability of survival. A frequently used modality for the screening and diagnosis of lung cancer is computed tomography (CT) imaging, which provides a detailed scan of the lung. In line with the advancement of computer-assisted systems, deep learning techniques have been extensively explored to help in interpreting the CT images for lung cancer identification. Hence, the goal of this review is to provide a detailed review of the deep learning techniques that were developed for screening and diagnosing lung cancer. This review covers an overview of deep learning (DL) techniques, the suggested DL techniques for lung cancer applications, and the novelties of the reviewed methods. This review focuses on two main methodologies of deep learning in screening and diagnosing lung cancer, which are classification and segmentation methodologies. The advantages and shortcomings of current deep learning models will also be discussed. The resultant analysis demonstrates that there is a significant potential for deep learning methods to provide precise and effective computer-assisted lung cancer screening and diagnosis using CT scans. At the end of this review, a list of potential future works regarding improving the application of deep learning is provided to spearhead the advancement of computer-assisted lung cancer diagnosis systems.

RESUMEN

Glaucoma is a chronic eye disease that may lead to permanent vision loss if it is not diagnosed and treated at an early stage. The disease originates from an irregular behavior in the drainage flow of the eye that eventually leads to an increase in intraocular pressure, which in the severe stage of the disease deteriorates the optic nerve head and leads to vision loss. Medical follow-ups to observe the retinal area are needed periodically by ophthalmologists, who require an extensive degree of skill and experience to interpret the results appropriately. To improve on this issue, algorithms based on deep learning techniques have been designed to screen and diagnose glaucoma based on retinal fundus image input and to analyze images of the optic nerve and retinal structures. Therefore, the objective of this paper is to provide a systematic analysis of 52 state-of-the-art relevant studies on the screening and diagnosis of glaucoma, which include a particular dataset used in the development of the algorithms, performance metrics, and modalities employed in each article. Furthermore, this review analyzes and evaluates the used methods and compares their strengths and weaknesses in an organized manner. It also explored a wide range of diagnostic procedures, such as image pre-processing, localization, classification, and segmentation. In conclusion, automated glaucoma diagnosis has shown considerable promise when deep learning algorithms are applied. Such algorithms could increase the accuracy and efficiency of glaucoma diagnosis in a better and faster manner.

RESUMEN

In ultrasound B-mode imaging, the axial resolution (AR) is commonly determined by the duration or bandwidth of an excitation signal. A shorter-duration pulse will produce better resolution compared to a longer one but with compromised penetration depth. Instead of relying on the pulse duration or bandwidth to improve the AR, an alternative method termed filtered multiply and sum (FMAS) has been introduced in our previous work. For spatial-compounding, FMAS uses the autocorrelation technique as used in filtered-delay multiply and sum (FDMAS), instead of conventional averaging. FMAS enables a higher frame rate and less computational complexity than conventional plane-wave compound imaging beamformed with delay and sum (DAS) and FDMAS. Moreover, it can provide an improved contrast ratio and AR. In previous work, no explanation was given on how FMAS was able to improve the AR. Thus, in this work, we discuss in detail the theory behind the proposed FMAS algorithm and how it is able to improve the spatial resolution mainly in the axial direction. Simulations, experimental phantom measurements and in vivo studies were conducted to benchmark the performance of the proposed method. We also demonstrate how the suggested new algorithm may be used in a practical biomedical imaging application. The balloon snake active contour segmentation technique was applied to the ultrasound B-mode image of a common carotid artery produced with FMAS. The suggested method is capable of reducing the number of iterations for the snake to settle on the region-of-interest contour, accelerating the segmentation process.

RESUMEN

In general, most of the existing convolutional neural network (CNN)-based deep-learning models suffer from spatial-information loss and inadequate feature-representation issues. This is due to their inability to capture multiscale-context information and the exclusion of semantic information throughout the pooling operations. In the early layers of a CNN, the network encodes simple semantic representations, such as edges and corners, while, in the latter part of the CNN, the network encodes more complex semantic features, such as complex geometric shapes. Theoretically, it is better for a CNN to extract features from different levels of semantic representation because tasks such as classification and segmentation work better when both simple and complex feature maps are utilized. Hence, it is also crucial to embed multiscale capability throughout the network so that the various scales of the features can be optimally captured to represent the intended task. Multiscale representation enables the network to fuse low-level and high-level features from a restricted receptive field to enhance the deep-model performance. The main novelty of this review is the comprehensive novel taxonomy of multiscale-deep-learning methods, which includes details of several architectures and their strengths that have been implemented in the existing works. Predominantly, multiscale approaches in deep-learning networks can be classed into two categories: multiscale feature learning and multiscale feature fusion. Multiscale feature learning refers to the method of deriving feature maps by examining kernels over several sizes to collect a larger range of relevant features and predict the input images' spatial mapping. Multiscale feature fusion uses features with different resolutions to find patterns over short and long distances, without a deep network. Additionally, several examples of the techniques are also discussed according to their applications in satellite imagery, medical imaging, agriculture, and industrial and manufacturing systems.

Asunto(s)

RESUMEN

Micro-expression analysis is the study of subtle and fleeting facial expressions that convey genuine human emotions. Since such expressions cannot be controlled, many believe that it is an excellent way to reveal a human's inner thoughts. Analyzing micro-expressions manually is a very time-consuming and complicated task, hence many researchers have incorporated deep learning techniques to produce a more efficient analysis system. However, the insufficient amount of micro-expression data has limited the network's ability to be fully optimized, as overfitting is likely to occur if a deeper network is utilized. In this paper, a complete deep learning-based micro-expression analysis system is introduced that covers the two main components of a general automated system: spotting and recognition, with also an additional element of synthetic data augmentation. For the spotting part, an optimized continuous labeling scheme is introduced to spot the apex frame in a video. Once the apex frames have been recognized, they are passed to the generative adversarial network to produce an additional set of augmented apex frames. Meanwhile, for the recognition part, a novel convolutional neural network, coined as Optimal Compact Network (OC-Net), is introduced for the purpose of emotion recognition. The proposed system achieved the best F1-score of 0.69 in categorizing the emotions with the highest accuracy of 79.14%. In addition, the generated synthetic data used in the training phase also contributed to performance improvement of at least 0.61% for all tested networks. Therefore, the proposed optimized and compact deep learning system is suitable for mobile-based micro-expression analysis to detect the genuine human emotions.

Asunto(s)

RESUMEN

Understanding a person's attitude or sentiment from their facial expressions has long been a straightforward task for humans. Numerous methods and techniques have been used to classify and interpret human emotions that are commonly communicated through facial expressions, with either macro- or micro-expressions. However, performing this task using computer-based techniques or algorithms has been proven to be extremely difficult, whereby it is a time-consuming task to annotate it manually. Compared to macro-expressions, micro-expressions manifest the real emotional cues of a human, which they try to suppress and hide. Different methods and algorithms for recognizing emotions using micro-expressions are examined in this research, and the results are presented in a comparative approach. The proposed technique is based on a multi-scale deep learning approach that aims to extract facial cues of various subjects under various conditions. Then, two popular multi-scale approaches are explored, Spatial Pyramid Pooling (SPP) and Atrous Spatial Pyramid Pooling (ASPP), which are then optimized to suit the purpose of emotion recognition using micro-expression cues. There are four new architectures introduced in this paper based on multi-layer multi-scale convolutional networks using both direct and waterfall network flows. The experimental results show that the ASPP module with waterfall network flow, which we coined as WASPP-Net, outperforms the state-of-the-art benchmark techniques with an accuracy of 80.5%. For future work, a high-resolution approach to multi-scale approaches can be explored to further improve the recognition performance.

Asunto(s)

RESUMEN

Pterygium is an eye condition that causes the fibrovascular tissues to grow towards the corneal region. At the early stage, it is not a harmful condition, except for slight discomfort for the patients. However, it will start to affect the eyesight of the patient once the tissues encroach towards the corneal region, with a more serious impact if it has grown into the pupil region. Therefore, this condition needs to be identified as early as possible to halt its growth, with the use of simple eye drops and sunglasses. One of the associated risk factors for this condition is a low educational level, which explains the reason that the majority of the patients are not aware of this condition. Hence, it is important to develop an automated pterygium screening system based on simple imaging modalities such as a mobile phone camera so that it can be assessed by many people. During the early stage of automated pterygium screening system development, conventional machine learning techniques such as support vector machines and artificial neural networks are the de facto algorithms to detect the presence of pterygium tissues. However, with the arrival of the deep learning era, coupled with the availability of large training data, deep learning networks have replaced the conventional networks in screening for the pterygium condition. The deep learning networks have been successfully implemented for three major purposes, which are to classify an image regarding whether there is the presence of pterygium tissues or not, to localize the lesion tissues through object detection methodology, and to semantically segment the lesion tissues at the pixel level. This review paper summarizes the type, severity, risk factors, and existing state-of-the-art technology in automated pterygium screening systems. A few available datasets are also discussed in this paper for both classification and segmentation tasks. In conclusion, a computer-assisted pterygium screening system will benefit many people all over the world, especially in alerting them to the possibility of having this condition so that preventive actions can be advised at an early stage.

RESUMEN

This study presents the efficiency of the wind-driven optimisation (WDO) approach in solving non-convex economic dispatch problems with point-valve effect. The best economic dispatch for a power system is one wherein the system can generate energy at a low cost. The calculation of the generating cost is subject to a number of constraints, such as the power demand for the entire system and the generation limit for each generator unit in the system. In addition, the system should also produce low power loss. The WDO optimisation technique is developed based on the concept of natural wind movement, which serves as a stabiliser to equalise the inequality of air pressure in the atmosphere. One major advantage of WDO over other techniques is its search accuracy. The proposed algorithm has been implemented in two systems, namely, the 10-generator and 40-generator systems. Both systems were tested in a Matlab environment. To highlight the capabilities of WDO, the results using this proposed technique are compared with the results obtained using flower pollination algorithm, moth flame optimisation, particle swarm optimisation and evolutionary programming techniques to determine the efficiency of the proposed approach in solving economic dispatch. The simulation results show the capability of WDO in determining the optimal power generation value with minimum generation cost and low rate of power loss.

RESUMEN

In recent years, there has been an immense amount of research into fall event detection. Generally, a fall event is defined as a situation in which a person unintentionally drops down onto a lower surface. It is crucial to detect the occurrence of fall events as early as possible so that any severe fall consequences can be minimized. Nonetheless, a fall event is a sporadic incidence that occurs seldomly that is falsely detected due to a wide range of fall conditions and situations. Therefore, an automated fall frame detection system, which is referred to as the SmartConvFall is proposed to detect the exact fall frame in a video sequence. It is crucial to know the exact fall frame as it dictates the response time of the system to administer an early treatment to reduce the fall's negative consequences and related injuries. Henceforth, searching for the optimal training configurations is imperative to ensure the main goal of the SmartConvFall is achieved. The proposed SmartConvFall consists of two parts, which are object tracking and instantaneous fall frame detection modules that rely on deep learning representations. The first stage will track the object of interest using a fully convolutional neural network (CNN) tracker. Various training configurations such as optimizer, learning rate, mini-batch size, number of training samples, and region of interest are individually evaluated to determine the best configuration to produce the best tracker model. Meanwhile, the second module goal is to determine the exact instantaneous fall frame by modeling the continuous object trajectories using the Long Short-Term Memory (LSTM) network. Similarly, the LSTM model will undergo various training configurations that cover different types of features selection and the number of stacked layers. The exact instantaneous fall frame is determined using an assumption that a large movement difference with respect to the ground level along the vertical axis can be observed if a fall incident happened. The proposed SmartConvFall is a novel technique as most of the existing methods still relying on detection rather than the tracking module. The SmartConvFall outperforms the state-of-the-art trackers, namely TCNN and MDNET-N trackers, with the highest expected average overlap, robustness, and reliability metrics of 0.1619, 0.6323, and 0.7958, respectively. The SmartConvFall also managed to produce the lowest number of tracking failures with only 43 occasions. Moreover, a three-stack LSTM delivers the lowest mean error with approximately one second delay time in locating the exact instantaneous fall frame. Therefore, the proposed SmartConvFall has demonstrated its potential and suitability to be implemented for a real-time application that could help to avoid any crucial fall consequences such as death and internal bleeding if the early treatment can be administered.

Asunto(s)

RESUMEN

Since the start of the COVID-19 pandemic at the end of 2019, more than 170 million patients have been infected with the virus that has resulted in more than 3.8 million deaths all over the world. This disease is easily spreadable from one person to another even with minimal contact, even more for the latest mutations that are more deadly than its predecessor. Hence, COVID-19 needs to be diagnosed as early as possible to minimize the risk of spreading among the community. However, the laboratory results on the approved diagnosis method by the World Health Organization, the reverse transcription-polymerase chain reaction test, takes around a day to be processed, where a longer period is observed in the developing countries. Therefore, a fast screening method that is based on existing facilities should be developed to complement this diagnosis test, so that a suspected patient can be isolated in a quarantine center. In line with this motivation, deep learning techniques were explored to provide an automated COVID-19 screening system based on X-ray imaging. This imaging modality is chosen because of its low-cost procedures that are widely available even in many small clinics. A new convolutional neural network (CNN) model is proposed instead of utilizing pre-trained networks of the existing models. The proposed network, Residual-Shuffle-Net, comprises four stacks of the residual-shuffle unit followed by a spatial pyramid pooling (SPP) unit. The architecture of the residual-shuffle unit follows an hourglass design with reduced convolution filter size in the middle layer, where a shuffle operation is performed right after the split branches have been concatenated back. Shuffle operation forces the network to learn multiple sets of features relationship across various channels instead of a set of global features. The SPP unit, which is placed at the end of the network, allows the model to learn multi-scale features that are crucial to distinguish between the COVID-19 and other types of pneumonia cases. The proposed network is benchmarked with 12 other state-of-the-art CNN models that have been designed and tuned specially for COVID-19 detection. The experimental results show that the Residual-Shuffle-Net produced the best performance in terms of accuracy and specificity metrics with 0.97390 and 0.98695, respectively. The model is also considered as a lightweight model with slightly more than 2 million parameters, which makes it suitable for mobile-based applications. For future work, an attention mechanism can be integrated to target certain regions of interest in the X-ray images that are deemed to be more informative for COVID-19 diagnosis.

RESUMEN

Pterygium is an eye condition that is prevalent among workers that are frequently exposed to sunlight radiation. However, most of them are not aware of this condition, which motivates many volunteers to set up health awareness booths to give them free health screening. As a result, a screening tool that can be operated on various platforms is needed to support the automated pterygium assessment. One of the crucial functions of this assessment is to extract the infected regions, which directly correlates with the severity levels. Hence, Group-PPM-Net is proposed by integrating a spatial pyramid pooling module (PPM) and group convolution to the deep learning segmentation network. The system uses a standard mobile phone camera input, which is then fed to a modified encoder-decoder convolutional neural network, inspired by a Fully Convolutional Dense Network that consists of a total of 11 dense blocks. A PPM is integrated into the network because of its multi-scale capability, which is useful for multi-scale tissue extraction. The shape of the tissues remains relatively constant, but the size will differ according to the severity levels. Moreover, group and shuffle convolution modules are also integrated at the decoder side of Group-PPM-Net by placing them at the starting layer of each dense block. The addition of these modules allows better correlation among the filters in each group, while the shuffle process increases channel variation that the filters can learn from. The results show that the proposed method obtains mean accuracy, mean intersection over union, Hausdorff distance, and Jaccard index performances of 0.9330, 0.8640, 11.5474, and 0.7966, respectively.

RESUMEN

Skeletal bone age assessment using X-ray images is a standard clinical procedure to detect any anomaly in bone growth among kids and babies. The assessed bone age indicates the actual level of growth, whereby a large discrepancy between the assessed and chronological age might point to a growth disorder. Hence, skeletal bone age assessment is used to screen the possibility of growth abnormalities, genetic problems, and endocrine disorders. Usually, the manual screening is assessed through X-ray images of the non-dominant hand using the Greulich-Pyle (GP) or Tanner-Whitehouse (TW) approach. The GP uses a standard hand atlas, which will be the reference point to predict the bone age of a patient, while the TW uses a scoring mechanism to assess the bone age using several regions of interest information. However, both approaches are heavily dependent on individual domain knowledge and expertise, which is prone to high bias in inter and intra-observer results. Hence, an automated bone age assessment system, which is referred to as Attention-Xception Network (AXNet) is proposed to automatically predict the bone age accurately. The proposed AXNet consists of two parts, which are image normalization and bone age regression modules. The image normalization module will transform each X-ray image into a standardized form so that the regressor network can be trained using better input images. This module will first extract the hand region from the background, which is then rotated to an upright position using the angle calculated from the four key-points of interest. Then, the masked and rotated hand image will be aligned such that it will be positioned in the middle of the image. Both of the masked and rotated images will be obtained through existing state-of-the-art deep learning methods. The last module will then predict the bone age through the Attention-Xception network that incorporates multiple layers of spatial-attention mechanism to emphasize the important features for more accurate bone age prediction. From the experimental results, the proposed AXNet achieves the lowest mean absolute error and mean squared error of 7.699 months and 108.869 months2, respectively. Therefore, the proposed AXNet has demonstrated its potential for practical clinical use with an error of less than one year to assist the experts or radiologists in evaluating the bone age objectively.

RESUMEN

BACKGROUND: Content-based medical image retrieval (CBMIR) system enables medical practitioners to perform fast diagnosis through quantitative assessment of the visual information of various modalities. METHODS: In this paper, a more robust CBMIR system that deals with both cervical and lumbar vertebrae irregularity is afforded. It comprises three main phases, namely modelling, indexing and retrieval of the vertebrae image. The main tasks in the modelling phase are to improve and enhance the visibility of the x-ray image for better segmentation results using active shape model (ASM). The segmented vertebral fractures are then characterized in the indexing phase using region-based fracture characterization (RB-FC) and contour-based fracture characterization (CB-FC). Upon a query, the characterized features are compared to the query image. Effectiveness of the retrieval phase is determined by its retrieval, thus, we propose an integration of the predictor model based cross validation neural network (PMCVNN) and similarity matching (SM) in this stage. The PMCVNN task is to identify the correct vertebral irregularity class through classification allowing the SM process to be more efficient. Retrieval performance between the proposed and the standard retrieval architectures are then compared using retrieval precision (Pr@M) and average group score (AGS) measures. RESULTS: Experimental results show that the new integrated retrieval architecture performs better than those of the standard CBMIR architecture with retrieval results of cervical (AGS > 87%) and lumbar (AGS > 82%) datasets. CONCLUSIONS: The proposed CBMIR architecture shows encouraging results with high Pr@M accuracy. As a result, images from the same visualization class are returned for further used by the medical personnel.

Asunto(s)

RESUMEN

Recycling is one of the most efficient methods for environmental friendly waste management. Among municipal wastes, plastics are the most common material that can be easily recycled and polyethylene terephthalate (PET) is one of its major types. PET material is used in consumer goods packaging such as drinking bottles, toiletry containers, food packaging and many more. Usually, a recycling process is tailored to a specific material for optimal purification and decontamination to obtain high grade recyclable material. The quantity and quality of the sorting process are limited by the capacity of human workers that suffer from fatigue and boredom. Several automated sorting systems have been proposed in the literature that include using chemical, proximity and vision sensors. The main advantages of vision based sensors are its environmentally friendly approach, non-intrusive detection and capability of high throughput. However, the existing methods rely heavily on deterministic approaches that make them less accurate as the variations in PET plastic waste appearance are too high. We proposed a probabilistic approach of modeling the PET material by analyzing the reflection region and its surrounding. Three parameters are modeled by Gaussian and exponential distributions: color, size and distance of the reflection region. The final classification is made through a supervised training method of likelihood ratio test. The main novelty of the proposed method is the probabilistic approach in integrating various PET material signatures that are contaminated by stains under constant lighting changes. The system is evaluated by using four performance metrics: precision, recall, accuracy and error. Our system performed the best in all evaluation metrics compared to the benchmark methods. The system can be further improved by fusing all neighborhood information in decision making and by implementing the system in a graphics processing unit for faster processing speed.

Asunto(s)

RESUMEN

Event recognition is one of the most active research areas in video surveillance fields. Advancement in event recognition systems mainly aims to provide convenience, safety and an efficient lifestyle for humanity. A precise, accurate and robust approach is necessary to enable event recognition systems to respond to sudden changes in various uncontrolled environments, such as the case of an emergency, physical threat and a fire or bomb alert. The performance of sudden event recognition systems depends heavily on the accuracy of low level processing, like detection, recognition, tracking and machine learning algorithms. This survey aims to detect and characterize a sudden event, which is a subset of an abnormal event in several video surveillance applications. This paper discusses the following in detail: (1) the importance of a sudden event over a general anomalous event; (2) frameworks used in sudden event recognition; (3) the requirements and comparative studies of a sudden event recognition system and (4) various decision-making approaches for sudden event recognition. The advantages and drawbacks of using 3D images from multiple cameras for real-time application are also discussed. The paper concludes with suggestions for future research directions in sudden event recognition.

Asunto(s)

RESUMEN

Telepointer is a powerful tool in the telemedicine system that enhances the effectiveness of long-distance communication. Telepointer has been tested in telemedicine, and has potential to a big influence in improving quality of health care, especially in the rural area. A telepointer system works by sending additional information in the form of gesture that can convey more accurate instruction or information. It leads to more effective communication, precise diagnosis, and better decision by means of discussion and consultation between the expert and the junior clinicians. However, there is no review paper yet on the state of the art of the telepointer in telemedicine. This paper is intended to give the readers an overview of recent advancement of telepointer technology as a support tool in telemedicine. There are four most popular modes of telepointer system, namely cursor, hand, laser and sketching pointer. The result shows that telepointer technology has a huge potential for wider acceptance in real life applications, there are needs for more improvement in the real time positioning accuracy. More results from actual test (real patient) need to be reported. We believe that by addressing these two issues, telepointer technology will be embraced widely by researchers and practitioners.

Asunto(s)

RESUMEN

In video analytics, robust observation detection is very important as the content of the videos varies a lot, especially for tracking implementation. Contrary to the image processing field, the problems of blurring, moderate deformation, low illumination surroundings, illumination change and homogenous texture are normally encountered in video analytics. Patch-Based Observation Detection (PBOD) is developed to improve detection robustness to complex scenes by fusing both feature- and template-based recognition methods. While we believe that feature-based detectors are more distinctive,however, for finding the matching between the frames are best achieved by a collection of points as in template-based detectors. Two methods of PBOD-the deterministic and probabilistic approaches-have been tested to find the best mode of detection. Both algorithms start by building comparison vectors at each detected points of interest. The vectors are matched to build candidate patches based on their respective coordination. For the deterministic method, patch matching is done in 2-level test where threshold-based position and size smoothing are applied to the patch with the highest correlation value. Forthe second approach, patch matching is done probabilistically by modelling the histograms of the patches by Poisson distributions for both RGB and HSV colour models. Then,maximum likelihood is applied for position smoothing while a Bayesian approach is appliedfor size smoothing. The result showed that probabilistic PBOD outperforms the deterministic approach with average distance error of 10.03% compared with 21.03%. This algorithm is best implemented as a complement to other simpler detection methods due to heavy processing requirement.

RESUMEN

Foreground detection has been used extensively in many applications such as people counting, traffic monitoring and face recognition. However, most of the existing detectors can only work under limited conditions. This happens because of the inability of the detector to distinguish foreground and background pixels, especially in complex situations. Our aim is to improve the robustness of foreground detection under sudden and gradual illumination change, colour similarity issue, moving background and shadow noise. Since it is hard to achieve robustness using a single model, we have combined several methods into an integrated system. The masked grey world algorithm is introduced to handle sudden illumination change. Colour co-occurrence modelling is then fused with the probabilistic edge-based background modelling. Colour co-occurrence modelling is good in filtering moving background and robust to gradual illumination change, while an edge-based modelling is used for solving a colour similarity problem. Finally, an extended conditional random field approach is used to filter out shadow and afterimage noise. Simulation results show that our algorithm performs better compared to the existing methods, which makes it suitable for higher-level applications.