ETD

• Artificial intelligence
• Deep learning
• Embedded system
• Light-weight models
• Surveillance applications
• video & Image processing

In the recent decade, artificial intelligence technology (AI technology) has entered an extraordinary phase of fast development and wide application. The techniques developed in traditional AI research areas, such as computer vision and object recognition, have found many innovative applications in an array of real-world settings. The general methodological contributions from AI, such as a variety of recently developed deep learning algorithms, have also been applied to a wide spectrum of fields such as surveillance applications, Real-time processing, IoT devices, and health care systems. The state-of-the-art and deep learning models have wider applicability and are highly efficient. The performance of these models on embedded systems is significantly limited by the platform’s CPU/GPU, memory, and power consumption, and their scope is limited to simplistic inference tasks only. The smaller the size of the trained model, the less storage and computing power is required to run the trained model on edge devices. In this PhD work, we present efficient implementations of different deep learning algorithms for image/video processing. These algorithms aim to identify and classify the specific objects relevant to each scenario of application. For each scenario,
we then select the algorithms that are most effective. This research focuses on the optimization of the deep learning models by designing light-weight architectures for embedded applications. This PhD work achieved the high accuracy needed while also being suitable for low-cost, low-performance hardware such as a Raspberry Pi or NVIDIA board. This PhD thesis is a collection of a sequential (published) papers in international journals and conferences to represent and construct each chapter of this thesis except the first and last one. The author of this thesis is the main author of these papers. This effort could not be beneficial or even done without getting a support from the colleagues, co-authors and supervisors who guide the author to develop his skills, II grew his scientific potential, and engage him in different research teams for exchanging ideas, concepts, and opinions. Furthermore, the practical and experiment activities require many specialists and technicians able to achieve the proposed concepts, design, and methodologies that the author developed.