Lijuan Yang,∗ Ying Li,∗ and Hui Zhao∗
Face detection, tracking algorithms, real-time video streams, double deep Q-network (DDQN), optimisation, surveillance. ∗ School of Computer Science and Engineering, North China Institute of Aerospace Engineering, Langfang 065000, China; e-mail: [email protected]; Ying [email protected]; ZhaoHui [email protected] Corresponding author: Lijuan Yang
Improving methods for detecting and tracking faces in live video feeds is essential for many applications, including surveillance and human–computer interaction. Nevertheless, the accuracy and efficiency of existing algorithms are compromised while trying to fulfil the demanding criteria of real-time processing. The computational constraints imposed by real-time processing requirements, lighting conditions, obstructions, and dynamic surroundings limit the performance of face detection and tracking systems. This research introduces a novel strategy for optimising face detection and tracking with the double deep Q-learning (FDT-DDQN) model in real-time video streams to enhance efficiency and accuracy in dynamic and resource-constrained scenarios. The FDT-DDQN model incorporates a reward structure that promotes timely and precise face identification, explores optimal network designs for double deep Q-network (DDQN) in face processing, and integrates pretreatment techniques to improve algorithmic robustness. The DDQN is selected due to its capacity to manage intricate decision- making tasks and acquire optimal policies from input data with high dimensions. Modern face recognition and tracking algorithms in live video feeds are far more precise and efficient due to the reinforcement learning framework. It reduces overestimation bias and stabilises the learning process. The project will conduct thorough experimental evaluations utilising benchmark datasets to evaluate and compare key performance parameters, including detection accuracy, processing rate, and memory use, against baseline implementations. The results are anticipated to showcase the efficacy of the suggested method in enhancing face detection and tracking techniques for live video feeds.
Important Links:
Go Back
