An energy efficient system for unmanned aerial vehicle-based surveillance

Smart surveillance has become an integral part of smart cities and urban living. Recently, unmanned aerial vehicle (UAV) or drone-based surveillance has become popular in smart cities for monitoring suspicious activities as well as for managing transport and emergency services. As the traditional surveillance systems operated by humans are constrained by human cognitive abilities and attention span, therefore a noticeable deterioration in object recognition can be witnessed. It is evident that human error can reach as high as 95% after about 22 minutes, which is significant for any catastrophic consequences. Additionally, automated surveillance systems, though effective, are demanding in terms of computing and memory resources. These computational and memory complexities lead to substantial energy consumption, making them inappropriate for remote surveillance, such as drone-based surveillance. In response to these challenges, we have engineered an adaptive system leveraging the deep learning architecture YOLO V7 in various quantization units. This has effectively reduced power usage by 1.6 times. This innovative system achieves a 22% decrease in graphics processing unit (GPU) consumption and reduces memory requirements by 29%, resulting in a surveillance system that outperforms traditional methods by being 1.43 times faster. This contributes to a more efficient solution for remote surveillance needs. This newly proposed system, with its significant reduction in power consumption, introduces a sustainable and robust surveillance alternative that retains its performance even in adverse conditions. The considerable speed boost facilitates a quick response, thereby making it a strong asset for maintaining security and swiftly addressing potential threats. Furthermore, we have evaluated this model's effectiveness for detecting suspicious objects, such as weapons, with an impressive accuracy rate of 93%. This makes it not only a more energy-efficient option but also an incredibly precise tool for identifying potential security threats.