Milad Ghantous

MIRF: A Multimodal Image Registration and Fusion Module Based on DT-CWT

Image Fusion is a powerful and necessary tool to incorporate the relevant visual information provided by multiple sensors simultaneously. The quality of the results however, is bounded not only by the quality of the algorithm, but also by the outcome of the required image registration algorithm. Despite this dependency, images are always assumed to be pre-aligned. With 3rd Generation surveillance systems, centralized computations are shifted to distributed visual nodes low on computational and power resources. This article presents a combined approach that is able to register and fuse multimodal images, dubbed MIRF. Combining both algorithms into one image domain not only offers a reduction in complexity making it a better fit for a resource constrained embedded platform, but also improves the response time of the system. Two algorithms for area-based image registration and object-based image fusion are proposed. They are based on Dual-Tree Complex Wavelet Transform. Qualitative and quantitative experimental results show that the proposed registration approach achieves comparable accuracies to its counterparts, with lower-complexity. On the other hand, the developed fusion scheme exhibits higher accuracy and proves its immunity to minor errors in registration

Visual Sensor Nodes

Recent developments in image and video technologies enabled easy access to a new type of sensor-based networks, Visual Sensor Networks (VSN). VSNs are gaining a lot of attention lately. They are used in several applications including surveillance and telepresence. They consist of several low-cost, low-power visual nodes with sensing, data processing, and communication capabilities. These tiny nodes are able to collect large volumes of images, process them, and send extracted data to each other and to the base station for further analysis. Unfortunately, the huge amount of data captured and processed is faced with the limited resources of such platforms. There are several challenges involved with the design and implementation of VSNs. This chapter presents an overview of visual nodes, architectures, and challenges. It also reviews available VSN platforms and compares their processing capabilities, highlighting the need for new lightweight but efficient image processing algorithms and architectures.

iSee: An Android Application for the Assistance of the Visually Impaired

Smart phone technology and mobile applications have become an indispensable part of our daily life. The primary use however, is targeted towards social media and photography. While some camera-based approaches provided partial solutions for the visually impaired, they still constitute a cumbersome process for the user. iSee is an Android based application that benefits from the commercially available technology to help the visually impaired people improve their day-to-day activities. A single screen tap in iSee is able to serve as a virtual eye by providing a sense of seeing to the blind person by audibly communicating the object(s) names and description. iSee employs efficient object recognition algorithms based on FAST and BRIEF. Implementation results are promising and allow iSee to constitute a basis for more advanced applications.

Pattern recognition of EMG signals: Towards adaptive control of robotic arms

Given the importance of hands as dexterous instruments to execute daily life tasks and the huge disability people suffer from when losing their limbs, this paper uses the electrical activity of the muscles (EMG) as control signals in a pattern recognition control system to manipulate the movement of a motorized 3D printed robotic arm. A system to acquire the electromyography signals is first designed and tested. Some features are then extracted from the EMG signal to build a support vector machine classification model. The obtained results indicate the fidelity of acquired signals, the efficiency of the exoskeleton, and the accuracy of the classification process in achieving a robust myoelectric control system. The final results compose a backbone for a future development of a robust classification system to fulfill a complete prototype.

Hardware Architecture Assist for Critical Components

Despite the advancements in both software and hardware, the majority of accurate image processing algorithms still contain some computational parts. These critical sections are viewed as bottleneck especially when real-time response is desired. Specialized hardware solutions are developed to accelerate critical low-level such operations. In this chapter, two hardware architectures are presented to complement image processing algorithms discussed in previous chapters. The first is a fast and compact ASIC architecture for hysteresis thresholding and object feature extraction. The second is an efficient hardware implementation for image decomposition based on Discrete Wavelet Transform. Both architectures exhibit higher performance than their software counterpart and hence help in alleviating the burden off the processing tasks.

For better energy consumption and management in future cellular networks

The success of wireless mobile networks is leading to an increase of power consumption in cellular networks caused by the massive growth in the number of base stations worldwide to serve the large cellular traffic demand. Hence, new methods and strategies are needed in the management of mobile networks in order to decrease energy consumption while maintaining reasonable costs and high quality of service for the users. Therefore, a new algorithm is proposed in this paper to help in reducing energy consumption of heterogeneous Long Term Evolution (LTE) network using Macro and Micro base stations. This algorithm will compute the energy consumption of different cell coverage configurations and choose the least power demanding scenario. Numerical simulations will prove the efficiency of our algorithm in comparison with traditional and existing cellular coverage schémas.