Computer Vision Performance Analysis for Smart Doorbell System With IoT and Edge Computing
Gaurang Raval, Shailesh Arya, Pankesh Patel, Sharada Valiveti, Riya Shah, Saurin Parikh · Advances in information security, privacy, and ethics book series · 2024
The Artificial Intelligence of Things (AIoT) includes machine learning applications, algorithms, hardware, and software. AIoT can be roughly classified into - vibration, voice, and vision. All of these have distinct workloads and demand scalable solutions. The focus of this work is on vision-based applications. The current offerings are expensive, inflexible, and exclusive. There is a trade-off between the precision and portability. To address these issues, a video analytics-based solution is proposed. It processes the smart doorbell data in real-time. The system is able to distinguish known/unknown people with high accuracy. It also detects animal/pet, harmful weapon, noteworthy vehicle, and package. Various approaches are applied for detection like cloud computing, IoT boards, classical computer vision. As part of this research, we wanted to collate contemporary video analytics with privacy, security, energy usage, and opacity with focus on Hardware/software cost, resource usage, accuracy, and latency. The approach best suitable for application development is thereby concluded.