Papers by ashley varghese
Advances in Intelligent Systems and Computing
The emergence of low-cost commercial drones fitted with a camera are ideal platforms for remotely... more The emergence of low-cost commercial drones fitted with a camera are ideal platforms for remotely monitoring critical assets such as railway corridor. The proposed system employs drones to automate and make the process efficient. In this paper, a railway monitoring system capable of detection and classification of various railway-related infrastructures such as lines, ballast, anchors, sleepers and fasteners using visual images captured by a drone is proposed. The first stage of classification uses a deep network that helps in qualifying the presence of track in a given frame. The second stage helps in classification of objects within a frame for further analysis. Two different deep architectures are used in classification of railway infrastructure—the first for offline analysis that uses transfer learning using a pre-trained GoogLeNet model and the second approach that uses a new architecture for embedded implementation. Transfer learning results in an overall f-score of 89%, and the new architecture results in an overall f-score of 81% with at least 10\(\times \) reduction in parameters.
2017 International Conference on Digital Image Computing: Techniques and Applications (DICTA), 2017
The use of autonomous drones in industrial inspection is gaining momentum with improvements in ha... more The use of autonomous drones in industrial inspection is gaining momentum with improvements in hardware and control. Considering the availability of historical data as drones gather information by regular sorties, a new opportunity for change detection is emerging for inspection and maintenance. In this paper, we propose a visual change detection framework using multi- scale super pixel approach. A framework that can (i) automatically capture appropriate images that match the stored images in the data agnostic drone store, and (ii) perform visual change detection is proposed. The algorithm is tested on a street-view change detection dataset containing images from 152 categories. Image matching step resulted in an accuracy of over 90%. On the street-view dataset, a change detection rate of over 90% is achieved with 60% categories detecting more than 60% of changed region. The proposed approach achieves better performance compared to other state-of-the- art methods that use image descriptors. The capability is implemented on an off-the-shelf drone to demonstrate the utility.
Communications in Computer and Information Science, 2020
Sparse coding techniques have shown to perform well in solving the conventional inverse imaging p... more Sparse coding techniques have shown to perform well in solving the conventional inverse imaging problems like inpainting and denoising. In the recent past, the performance of deep learning architectures in solving these inverse imaging problems have exceeded the conventional approaches several times. The only limitation of these architectures is the requirement of large volumes of training data. Deep dictionary learning (DDL) is an emerging approach and has been shown to solve some important classification problems in the scenarios where there is a scarcity of training data. DDL framework effectively combines the advantages of sparse coding and deep learning. In this paper, DDL framework is adapted to solve the inverse imaging problem with specific focus on imaging inpainting. An alternating minimization (AM) approach is proposed to derive the dictionaries and their corresponding sparse coefficients at each level of the DDL framework. The aim of this work is to show that the multilevel dictionaries can be leveraged to derive the sparse representations without compromising on the restoration quality of streaked multispectral images. Inspite of being a conventional machine learning based technique, we show that the performance of our approach is better to the state-of-the-art deep learning approaches for multispectral image inpainting.
2017 Fifteenth IAPR International Conference on Machine Vision Applications (MVA), 2017
The use of drones in infrastructure monitoring aims at decreasing the human effort and in achievi... more The use of drones in infrastructure monitoring aims at decreasing the human effort and in achieving consistency. Accurate aerial image analysis is the key block to achieve the same. Reliable detection and integrity checking of power line conductors in a diverse background are the most challenging in drone based automatic infrastructure monitoring. Most techniques in literature use first principle approach that tries to represent the image as features of interest. This paper proposes a machine learning approach for power line detection. A new deep learning architecture is proposed with very good results and is compared with GoogleNet pre-trained model. The proposed architecture uses Histogram of Gradient features as the input instead of the image itself to ensure capture of accurate line features. The system is tested on aerial image collected using drone. A healthy F-score of 84.6% is obtained using the proposed architecture as against 81% using GoogleNet model.
Lecture Notes in Computer Science, 2019
The increasing urban population in cities necessitates the need for the development of smart citi... more The increasing urban population in cities necessitates the need for the development of smart cities that can offer better services to its citizens. Drone technology plays a crucial role in the smart city environment and is already involved in a number of functions in smart cities such as traffic control and construction monitoring. A major challenge in fast growing cities is the encroachment of public spaces. A robotic solution using visual change detection can be used for such purposes. For the detection of encroachment, a drone can monitor outdoor urban areas over a period of time to infer the visual changes. Visual change detection is a higher level inference task that aims at accurately identifying variations between a reference image (historical) and a new test image depicting the current scenario. In case of images, the challenges are complex considering the variations caused by environmental conditions that are actually unchanged events. Human mind interprets the change by comparing the current status with historical data at intelligence level rather than using only visual information. In this paper, we present a deep architecture called ChangeNet for detecting changes between pairs of images and express the same semantically (label the change). A parallel deep convolutional neural network (CNN) architecture for localizing and identifying the changes between image pair has been proposed in this paper. The architecture is evaluated with VL-CMU-CD street view change detection, TSUNAMI and Google Street View (GSV) datasets that resemble drone captured images. The performance of the model for different lighting and seasonal conditions are experimented quantitatively and qualitatively. The result shows that ChangeNet outperforms the state of the art by achieving 98.3% pixel accuracy, 77.35% object based Intersection over Union (IoU) and 88.9% area under Receiver Operating Characteristics (RoC) curve.
2017 International Joint Conference on Neural Networks (IJCNN), 2017
Infrastructure detection and monitoring is a difficult task. Due to the advances in unmanned vehi... more Infrastructure detection and monitoring is a difficult task. Due to the advances in unmanned vehicles and image analytics, it is possible to decrease the human effort and achieve consistent results in infrastructure assessments using aerial image processing. Reliable detection and integrity checking of power infrastructure including conductor lines, pylons and insulators in a diverse background is the most challenging task in drone based automatic infrastructure monitoring. Most techniques in literature use first principle approach that tries to represent the image as features of interest. This paper proposes a deep learning approach for power infrastructure detection. Graph based post processing is applied for improving the outcomes of the generated deep model. A f-score of 75% is achieved using the deep model which is further improved using spectral clustering for the conductor lines, pylons and insulators that form the core parts of power infrastructure.
2016 30th International Conference on Advanced Information Networking and Applications Workshops (WAINA), 2016
In this era of digitization, there are many application swhich can be used for sharing videos to ... more In this era of digitization, there are many application swhich can be used for sharing videos to a group of people around the world in real time. During the acquisition of real time data, many sensitive information also gets captured along. This poses a threat to privacy of data. In this paper, we consider the scenario of real time sharing of video from multiple sources to multiple end users through a secure publish subscribe architecture. We propose a system that provides a secure user profile based data privacy using Hierarchical Inner Product Encryption (HIPE) and a broker with anonymous pub sub architecture. System provides multiple levels of data access control to end users based on their roles so that they will be able to see only those videos for which they have access. In order to ensure secure communication, the broker publishes data on encrypted topic and uses HIPE for data subscription. In this paper, we demonstrate the end-to-end working of the system through prototype implementation. Further, we also present the system performance and evaluate the system security both with and without HIPE.
2015 3rd International Conference on Signal Processing, Communication and Networking (ICSCN), 2015
Mobile Augmented Reality (MAR) is an emerging field with a wide variety of applications involving... more Mobile Augmented Reality (MAR) is an emerging field with a wide variety of applications involving the intricate challenges of computer vision. However, Mobile Augmented Reality for Inspection and Tele-Assistance (MARITA) is a relatively unknown entity which opens up new challenges in the field of wireless communications along with the existing computer vision hurdles. MARITA aims at providing augmented reality based solution for real-time assistance to a person in a remote area. The existing 3G technologies although good, do not provide enough up-link bandwidth to facilitate real-time video transmission. The time varying communication channel adds to the problem by introducing errors resulting in packet drops. Thus achieving reliable communication is a challenge. Error correction codes are often used to counter this problem. However, real-time constraints in video communication restrict the use powerful codes which may introduce considerable delay in transmission because of the complex encoding or decoding process. This paper addresses the issue by using appropriate error correction codes such as Reed-Solomon (RS) codes and Low Density Parity Check (LDPC) codes. A set LDPC codes considered have been obtained form the idea of Projective Geometry. In this paper, we present results obtained by applying these codes for video transmission in real-time achieved by our prototype implementation.
2015 Asia Pacific Conference on Multimedia and Broadcasting, 2015
At the outset, Augmented Reality (AR) technology in mobile applications is heavily leveraged. Mob... more At the outset, Augmented Reality (AR) technology in mobile applications is heavily leveraged. Mobile AR, as it is called is gaining popularity across all domains. This makes content security a prime area of concern. Hence it is inherent that the security and privacy concerns need to be addressed. In this paper, we propose "SMART" (Secure Mobile Augmented Reality for Tele-Assistance), a system which addresses the security and privacy issues associated with MART (Mobile Augmented Reality for Tele-Assistance) applications. We address the problem of image privacy which is a major concern in AR. This issue is addressed by using Attribute Based Encryption (ABE) method, which ensures a user profile based privacy. The identified privacy regions are encrypted using Advanced Encryption Standard (AES) and the AES keys are protected using ABE. Only the user with a valid access profile will be able to decrypt the image. A Pub-Sub architecture is used to communicate between the field and end users on UDP connection through a broker using MQTT protocol. In this paper, we demonstrate the end-to-end working of the system through prototype implementation. The performance of the system is also evaluated on the criteria of latency and packet size.
2015 Asia Pacific Conference on Multimedia and Broadcasting, 2015
Tele-assistance facilitated by Mobile Augmented Reality (MAR) has been suggested as a powerful wa... more Tele-assistance facilitated by Mobile Augmented Reality (MAR) has been suggested as a powerful way to interact and collaborate while providing remote services. In such scenarios, one has to cater for uncontrolled environments with little or no prior information about the scene is available. This poses significant challenges towards rendering accurate augmented graphics. In this paper, we propose a framework, where the necessary scene information extracted from a single image of the scene, is used for tracking and augmenting the graphics. We have focused on the use of deformable models for object detection and tracking with a particular case of vehicles as objects of interest. The proposed framework is demonstrated with the help of a fully functioning prototype for interaction between user and remote expert. Apart from providing the relevant technical details, the paper also presents some promising and useful results achieved during the process.
2015 IEEE 9th International Symposium on Intelligent Signal Processing (WISP) Proceedings, 2015
Mobile Augmented Reality (MAR) is an emerging field and its nascent applications are finding its ... more Mobile Augmented Reality (MAR) is an emerging field and its nascent applications are finding its ways into the current deployments of cyber physical system. Mobile devices can harness augmented reality technology in any unprepared environment. This introduces a challenge to achieve an accurate and robust registration and tracking of mobile device. For accurate tracking, much research is being carried out to fuse inertial and vision sensor data. The resultant tracking can be further made better by finding means to track coupled translational and rotational motions. This problem is tackled with a neat formalism in terms of dual quaternion. Unit dual quaternion can capture the coupling between translational and rotational motions. In this paper, the requisite machinery is pivoted around Extended Kalman Filter (EKF) and is derived based on dual quaternion. The derived EKF expression is verified through experimentation involving both simulated and realistic data, the latter being obtained from a prototype for MAR. The simulation results show the effectiveness of dual quaternion on position and orientation estimation. This novel fusion framework resulted in more accurate tracking as compared to that of the existing quaternion based algorithm.
2014 IEEE International Conference on Advanced Networks and Telecommuncations Systems (ANTS), 2014
MobiCoStream, an acronym for Mobile Collaborative upStream, refers to Real-time video upstream of... more MobiCoStream, an acronym for Mobile Collaborative upStream, refers to Real-time video upstream of captured video in Mobile Augmented Reality (MAR) applications. MAR for tele-assistance deals with providing real-time assistance to the user through the power of Augmented Reality. Real-time applications have strict delay requirements which need to be met in order to satisfy the user Quality of Service (QoS). Mobile communication still being in its third generation, limits the availability of high up-link bandwidth for video transmission. The problem is tackled by means of Bandwidth Aggregation through which the hand-held devices in the vicinity, collaborate with the user to achieve real-time video up-link and maintain the user QoS. In this paper, we present the actual results pertaining to the prototype implementation of the proposed system. Also presented are some simulation results, by viewing the multiple hand-held devices as a virtual Multiple Input Multiple Output (MIMO) system and in turn considering the popular Alamouti scheme in this context.
The International journal of Multimedia & Its Applications, 2014
Mobile Augmented Reality (MAR) is becoming an important cyber-physical system application given t... more Mobile Augmented Reality (MAR) is becoming an important cyber-physical system application given the ubiquitous availability of mobile phones. With the need to operate in unprepared environments, accurate and robust registration and tracking has become an important research problem to solve. In fact, when MAR is used for tele-interactive applications involving large distances, say from an accident site to insurance office, tracking at both the ends is desirable and further it is essential to appropriately fuse inertial and vision sensors' data. In this paper, we present results and discuss some insights gained in marker-less tracking during the development of a prototype pertaining to an example use case related to breakdown/damage assessment of a vehicle. The novelty of this paper is in bringing together different components and modules with appropriate enhancements towards a complete working system.
Uploads
Papers by ashley varghese