Edge AI for Real-Time Object Tracking in Dynamic Environments using Hybrid Motion Estimation Models

In applications such as autonomous vehicles, surveillance, they need real-time object tracking in a dynamic environment. However, traditional motion estimation techniques are not accurate and are not easy to adapt, while deep learning-based models usually need high computational power, which is not a suitable thing for edge devices. This paper therefore proposes an Edge AI driven Hybrid Motion Estimation Model (H-MEM) that comprises DSLF, AKF, and OFE which can provide real-time and low latency object tracking. Finally, the model is deployed on the edge devices with AI accelerators for computational efficiency with high tracking accuracy. A hierarchical processing framework that works out how to balance workloads between the edge and the cloud can use resources that are cheap and plentiful. The proposed system is equipped with a self-learning mechanism through reinforcement learning to adapt to environmental variations like changes in lighting, occlusions, varying motion speeds, etc. The accuracy of tracking is shown to improve and latency to reduce as well as computational efficiency compared to prevalent models in experimental evaluations on standard datasets. This shows the potential applicability of the model for real-world applications with a decent robust and scalable solution to dynamic object tracking in the real world. With this, it enables the next generation intelligent tracking system by integrating AI with edge computing.

Перевод пока недоступен