Design and Optimization of AI-Driven Image Processing Pipelines for Smart Cameras

Huayapa B.V.; Quispe R.C.; Talavera J.S. · International Journal of Engineering Trends and Technology (2025) · DOI: 10.14445/22315381/ijett-v73i5p104

Resumen

This paper presents a new design and optimization framework for the AI-driven image-processing pipeline dedicated to smart cameras that overcome main computational efficiency, energy consumption, and flexibility challenges. The proposed pipeline integrates state-of-the-art deep learning models, such as YOLOv5 and Faster R-CNN, along with optimization techniques, including model quantization and pruning. The modular architecture is flexible for the integration of new algorithms and technologies. Extensive validation was made in urban traffic surveillance and residential security contexts in Arequipa, Peru, which shows significant improvements in accuracy, processing speed, and energy efficiency. The deployed pipeline achieves, on average, 92% precision, 89% recall rate, and 90.5% F1 score on the urban traffic monitoring domain while improving the processing speed by 40% and reducing energy consumption by 35%. For intrusion detection in home security, it detects with 88% accuracy, FPR of 5%, and FNR of 7%. Due to the modular nature of this design, it reduces the integration time of new functionalities by 60%. These results give reason for the robustness and feasibility of the pipeline that can be deployed in resource-constrained environments, opening perspectives toward wider diffusion and further research on AI-driven smart cameras.

Instituciones: Universidad Nacional de San Agustín de Arequipa