MUZAMMIL KHAN
@utwente.nl
Researcher in the Robotics and Mechatronics (RAM) Group at the Faculty of EEMCS, University of Twente, Enschede
University of Twente
Muzammil Khan is a researcher at the Faculty of EEMCS, University of Twente. His current research is focused on implementing simultaneous localization and mapping (SLAM) algorithms for improving laparoscopic liver resection. Prior to this, he was a Ph.D. scholar in the Department of Mathematics, Bioinformatics, and Computer Applications at the National Institute of Technology Bhopal, India. His broad research area encompasses modern computer vision and machine learning applications in the field of healthcare. He explored the topic of Optical Flow in his thesis titled “Novel Algorithms for Optical Flow Estimation and Its Application”. In his work, he developed mathematical models for accurate optical flow estimation and further utilized them with deep learning techniques to perform anomaly detection in image sequences. During his Ph.D., he wrote 7 journal articles, 11 international conferences, and 1 book chapter. He presented his research at 8 international conferences.
EDUCATION
Bachelor of Science (2016) from Bundelkhand University Jhansi
Master of Science (2018) from Bundelkhand University Jhansi
Ph.D. (since 2019 - expected completion in 2023) from Maulana Azad National Institute of Technology
CSIR-NET exam (2019) in Mathematical Sciences (qualified)
GATE (Graduate Aptitude Test in Engineering- 2019) exam in Mathematics (Qualified)
RESEARCH, TEACHING, or OTHER INTERESTS
Computer Vision and Pattern Recognition, Medical Laboratory Technology, Applied Mathematics
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
Scopus Publications
Muzammil Khan, Nitish Kumar Mahala, and Pushpendra Kumar
Springer Science and Business Media LLC
Muzammil Khan and Pushpendra Kumar
Elsevier BV
Muzammil Khan, Nitish Kumar Mahala, and Pushpendra Kumar
IEEE
As we are aware that many indoor and outdoor fires abruptly occur every day around the world, which results as a major factor to global warming, hundreds of casualties and serious threat to property safety. Thus, the early prediction of fire is particularly become our utmost priority to reduce the loss, because once the fire spreads, it is difficult to control. Generally, the early detection of fire has its association with raising smoke, which is the smoldering stage of fire. In this stage, the smoke demonstrates different textures, colors, and shapes. The novel idea of this paper is to perform the fire prediction using the statical and dynamical smoke features. The static features of smoke are considered as color, texture and shape, while the dynamic features are derived from its motion field. The smoke dynamic motion is computed in terms of optical flow field color map, and used to represent the active regions of the image sequence (video). Therefore, a fractional order based variational model is presented for optical flow determination. These estimated color maps are further divided into the three color channels. The color channels manifesting smoke motion sensitivity are segmented with the aid of binary masks. Next, the segmented color maps along with the corresponding image sequences are used in a novel CNN architecture to detect smoke. In order to validate the model performance, a thorough comparison study is also depicted with several state-of-the-art models. A diverse collection of datasets comprised of 16 and 19 smoke and non-smoke videos, respectively, is considered for experiments.
Bhavana Singh, Muzammil Khan, and Pushpendra Kumar
IEEE
The world entered into a new phase in the wake of Covid-19, which is declared as a global threat because countless people were losing their lives worldwide due to this disease every day. Even the best medical facilities were crumbling due to the rise of COVID-19 and lack of funding. However the use of nucleic acid-based testing generates an excessive amount of bio-waste and increases the risk of infection to the healthcare personnel. These tests take a lot of time, which contributes to the high mortality rate. To address these challenges, vision based techniques have been demonstrated to be beneficial in the early and swift diagnosis of COVID-19. Earlier imaging techniques used X-ray or CT-scan images to detect COVID-19. Consequently, these techniques do not filter out non-COVID cases, resulting in longer processing time. In this paper, COVID-19 identification is accomplished through the dynamic characteristics obtained from chest X-ray (CXR) images. These dynamic characteristics are derived as optical flow (OF). Moreover a fractional order-based variational model is used to estimate OF from two consecutive images of lungs. The novelty of this work is to use OF instead of images in COVID-19 detection. Further, the classification of lung dynamic features is performed with a convolutional neural network (CNN) in coordination of various machine learning (ML) classifiers. A detailed comparison is provided to show the significance of the algorithm.
Muzammil Khan and Pushpendra Kumar
IEEE
As we are aware that the world witnesses a huge number of fire breakouts everyday, which results in high numbers of hazardous events and severe losses to property and forest vegetation. Therefore, early stage fire detection is of vital importance, for once it spreads it becomes unmanageable and disastrous. The early detection of fire can be performed with the help of vision based deep learning techniques. The novelty of the work lies in performing the fire detection using the static and dynamic features of fire. The static fire features are taken as shape, texture, and color, while the dynamic feature accounts for its flickering motion. For this purpose, the fire motion is estimated in terms of optical flow from videos (image sequences) by using a motion edge preserving level set segmentation based fractional order variational model. Level sets provide nicely segmented flow fields, while fractional order derivatives are capable to deal with discontinuities in the motion field. The estimated optical flow field is used to derive four fire features, which are constituted as 4D vectors. These 4D vectors reduce the data dimensionality and mitigates over-fitting problem. Finally, the fire detection is carried out by implementing a mixed data CNN-LSTM model. The mixed data presented in the work is composed of a reference image frame and the corresponding 4D vector sequence. Also, the significance of the model is manifested through an ablation study. The model performance validation is performed thorough a comparison study conducted with several existing models.
Nitish Kumar Mahala, Muzammil Khan, and Pushpendra Kumar
IEEE
As we are aware that thousands of fires break out every day around the world, which results in high numbers of casualties and serious threat to property safety and forest vegetation. Hence, it becomes particularly important to detect the fire at its early stage, because once the fire has spread in an area, it gets cataclysmic and difficult to control. In particular, the early detection of fire is associated with rising smoke. Therefore, the smoke can be considered as a good indicator to predict fire. In the presented work, smoke detection is performed with the help of its dynamical features. The dynamical features are considered in the form of optical flow color map. The motivation of this work is to use fractional order optical flow instead of images to provide the precise location and rate of growth. The estimation of optical flow is carried out using a non-local weighted fractional order variational model, which is capable in preserving dynamical discontinuities in the optical flow. Optical flow helps to find the active region of the images (video). This non-local weight also incorporates the robustness against noise. Further, the optical flow field is converted into a color map using an RGB color wheel. These color maps are used in different deep learning models for training and testing. The experiments are conducted on a dataset consisting of 18 smoke and 17 non-smoke videos. Different accuracy metrics are used for performance evaluation and detailed comparisons in order to demonstrate the significance of optical flow color maps over images in smoke detection.
Pushpendra Kumar, Muzammil Khan, and Nitish Kumar Mahala
Springer Nature Switzerland
Muzammil Khan and Pushpendra Kumar
Elsevier BV
Pushpendra Kumar and Muzammil Khan
IOP Publishing
Abstract As we are aware that motion estimation is an active and challenging area of vision system, which leads to the applications of computer vision. In general, motion detection and tracking in the image sequence (video) is carried out based on optical flow. In the recent-past, researchers have made a significant contribution to the estimation of optical flow through integer order-based variational models, but these are limited to integer order differentiation. In this paper, a nonlinear modeling of fractional order variational model in optical flow estimation is introduced using the Charbonnier norm, which can be scaled to integer order L 1-norm. In particular, the variational functional is formulated by the inclusion of a non-quadratic penalty term, regularization term and the Marchaud’s fractional derivative. This non-quadratic penalty provides effective robustness against outliers, whereas the Marchaud’s fractional derivative possesses a non-local character, and therefore is capable to deal with discontinuous information about texture and edges, and yields a dense flow field. The numerical discretization of the Marchaud’s fractional derivative is employed with the help of Grünwald–Letnikov fractional derivative. The resulting nonlinear system is converted into a linear system and solved by an efficient numerical technique. Several experiments are performed over a spectrum of datasets. The robustness and accuracy of the proposed model are shown under different amounts of noise and through detailed comparisons with the recently published works.
Pushpendra Kumar, Muzammil Khan, and Shreya Gupta
IEEE
Due to the wide range of applications, video surveillance is known as one of the challenging tasks of computer vision which requires detecting and tracking the moving objects in a sequence of images (video). As we are aware that several environmental conditions such as fog, darkness, snow-fall, illumination, rain degrade the quality of vision system. This motivates us to develop a robust infrared (IR) surveillance system to fulfill the open-ended goals of the vision problem. The active motion region is detected by using optical flow. In this paper, an energy functional has been presented for optical flow estimation by incorporating the fractional order total variational (TV) and quadratic terms. In particular, the proposed model is convex and more robust against outliers and provides a dense flow. However, the total variation regularization term is of non-differentiable nature which makes the minimization scheme apparently difficult. The fractional derivative discretization of non-differentiable terms is performed by using Grunwald-Letnikov (GL) derivative. The Primal-dual algorithm is applied in solving the resulting minimization scheme. Finally, the resulting variational formulation is solved by using an appropriate method. The validity, efficiency, and robustness of the proposed system are tested on a variety of datasets under various conditions.
Pushpendra Kumar and Muzammil Khan
IEEE
Novel Coronavirus disease (COVID-19) is an infectious disease that has been declared as a pandemic by the World Health Organization. Both symptomatic, as well as asymptomatic patients, are equally likely capable of spreading the virus among the population. Therefore, a real-time tracing of COVID-19 suspects and their identification by a computer-based algorithm is a need of the current times, so that the spreaders could be isolated and the mushrooming should be halted. In this paper, we introduced a fractional order variational model for the early prediction and detection of COVID-19 suspects based on the CXR image sequence. The identification is performed in terms of optical flow color map. The proposed technique would be financially cheaper, require less time and manpower in comparison to the available techniques. The presented model keeps discontinuous information about texture and edges and offers a dense flow field for minuscule variations. The Grünwald-Letnikov derivative is employed for discretizing the complex fractional order partial derivatives. The validity of the model is verified through a variety of experimental results on various datasets.
Muzammil Khan and Pushpendra Kumar
IEEE
As we are aware that underwater motion estimation is an active and challenging area of vision system which belongs to the category of robot navigation. In particular, motion detection and tracking in underwater image sequences is carried out based on optical flow. In this paper, a vision based underwater navigation system is developed based on the fractional order total variation (TV) model. The objective of this work is to design a variational model by using a quadratic data and total variation terms to provide the optimal performance against radiometric characteristics such as turbidity, non-uniform illumination and marine snow, etc. The presented ameliorated model is more robust against outliers and reduces the problem of local minima. The fractional derivative discretization of non-differentiable terms is performed using Grünwald-Letnikov derivative scheme. Finally, the resulting variational formulation is solved by using an appropriate method. The validity, efficiency, and robustness of the proposed model are tested on a variety of datasets.
Muzammil Khan and Pushpendra Kumar
IEEE
Variational models are more popular approaches in the estimation of optical flow between two image frames and yield the most accurate flow fields. More fidelity terms in the variational model makes the estimation robust. This paper proposed an anisotropic operator which is designed using the greatest integer and an exponential function to estimate average flow velocity. This will help to preserve discontinuity in the optical flow and provides a significant smooth flow over a uniform region. The design of this operator is motivated from an isotropic operator that is based on the intensity differences of the pixels. This is employed in the controlling of flow propagation. The validation of the accuracy and robustness of our algorithm is provided in terms of qualitative and quantitative results on a variety of spectrum datasets.