| Title |
Research on Domain Adaptive Video Surveillance System |
| DOI |
https://doi.org/10.5573/ieie.2023.60.3.80 |
| Keywords |
AI camera; CNN; Intelligent video surveillance; Domain adaptation |
| Abstract |
There are various computer vision algorithms based on artificial intelligence (AI) have been applied on commercial solution in these days. However, it requires a high complexity using a graphic processing unit, so that it is really difficult to find commercialized AI services based on computer vision applications. In particular, AI cameras are still limited in the low-complexity and high accuracy convolutional neural networks (CNNs) model, and more importantly, do not easily design a robust system architecture between edge device and (cloud) server for real-world applications. Towards addressing these limitations, we propose a novel hybrid video surveillance system for robustly detecting objects, consisting of front-end and back-end intelligence. For intelligent front-end, we propose an optimized person detector for AI camera and we also develop a new domain adaptation method in order to update personal AI model of each camera by understanding the space and context information in real-time for intelligent back-end system. It is an iterative and continuous process that new upcoming data and previous model are engaged in a continual process of improvement consistently. We conducted a series of experiments showing high accuracy and versatility of the new architecture, while yielding robust results that can be practically implemented. |