Energy management strategies in heterogeneous computing platform for surveillance applications

Abstract

In the digital era, there is an increasing demand for advanced surveillance systems across diverse sectors such as security, environmental monitoring, and urban management which typically rely on a network of high-resolution cameras, radars, and sensors that provide continuous monitoring and data transmission to centralized stations. However, these surveillance systems face a significant challenge in managing continuous and demanding tasks like video capture, real-time monitoring, and data transmission due to their continuous operation and high processing demand which leads to significant power consumption and heat generation. High power consumption further creates several issues, including thermal management challenges, performance degradation of tasks, higher operating and cooling costs, reliability problems, regulatory constraints, etc. Thus, energy management in surveillance systems, especially for those deployed in remote or hard-to-access locations where battery-operated systems are used, poses significant operational challenges that require a combination of power-efficient hardware, smart energy management strategies, and innovative power sources, such as renewable energy or hybrid solutions. Thus, this thesis aims to provide an overall comprehensive solution to the energy management problem in surveillance systems by developing advanced software algorithms, and power-saving technologies, making use of efficient hardware, and integrating the system with renewable energy solutions for better efficiency, reduced operational costs, longer battery life, and minimized environmental impact which ensures that the systems remain reliable and scalable, even in challenging environments or large-scale deployments. The first group of research begins by identifying and addressing the vulnerabilities in existing energy management strategies with the development of a dynamic resource allocation framework. This framework is specifically designed for heterogenous multicore processor architecture and aims to dynamically allocate the resources based on the distinct characteristics of multimedia and non-multimedia tasks, as well as the current system state. By intelligently distributing computational power, the algorithm ensures that high-performance tasks receive adequate resources without unnecessarily increasing energy consumption. This optimization not only improves the overall performance of the surveillance system but also lays the foundation for effective thermal management by reducing the likelihood of excessive heat generation during resource-intensive operations. Building upon the energy-efficient resource allocation, the second group of research introduces a thermal phase-aware adaptive task migration framework. As surveillance tasks often involve intensive computation that can generate significant heat, this framework dynamically monitors the thermal state of the processor to detect potential overheating. By leveraging thermal phase detection, the system can predict thermal hotspots and initiate task migration, redistributing workloads across different cores or processors to balance thermal loads. This proactive thermal management ensures even heat dissipation, prevents performance degradation due to thermal throttling, and enhances the reliability and longevity of the hardware. The adaptive nature of this framework is closely linked to the resource allocation algorithm, as the system can adjust task distribution based on both energy consumption and thermal conditions, ensuring optimal performance and thermal stability. Finally, a cost-effective and energy-efficient surveillance system i.e. biodiversity sensor (BS) is developed which can track the movement of flying insects in real-time, along with environmental conditions and updates to the cloud server. BS is an IoT device, developed on Apalis iMX8 QuadMax which is a heterogenous octa-core processor and also includes dual GC700 GPUs for better performance, efficiency and extended battery life. It is powered by a solar-PV/battery bank and is enabled with a power management integrated circuit for power conservation. The preinstalled operating system (OS) is replaced with a customized Linux OS built with Yocto, and loaded with both a dynamic resource allocation algorithm and thermal phase-aware adaptive task migration for energy management. It also features an over-the-air (OTA) update capability for remote management and a device management portal for local control. The developed sensor has been tested in the field, and the result shows that the BS effectively captures the frames of flying insects and performs surveillance appropriately with an accuracy of more than 90%. Results also demonstrate superior power efficiency and CPU utilization compared to pre-installed multimedia OS alternatives, making it a robust and sustainable solution for effective biodiversity monitoring.