Energy-Efficient CNN Hardware Accelerator for Real-Time Construction Progress Monitoring

Real-time construction progress monitoring is a critical but resource-intensive task for autonomous edge devices. Traditional deep learning approaches face limitations in terms of computational power and energy consumption, which hinder their on-site deployment. This article introduces an innovative method to enhance the autonomy of edge devices by co-designing a customized Convolutional Neural Network (CNN) with a dedicated hardware accelerator. This integrated approach addresses the inherent trade-offs between model accuracy, computational efficiency, and energy consumption. Our proposed solution focuses on wall state classification for monitoring construction progress. We present a lightweight CNN architecture optimized for efficient hardware implementation, minimizing the need for complex, resource-heavy operations. The model is deployed on a custom-designed hardware accelerator, specifically tailored to execute the proposed CNN with high performance and minimal power draw. The evaluation demonstrates that our method achieves a 171× energy saving and a significant reduction in computational time compared to traditional software-based solutions. This co-design methodology contributes to the body of knowledge by presenting a practical and energy-efficient solution for real-time AI applications in civil construction. Our approach provides a clear path for deploying machine learning models on resource-constrained edge devices, thereby advancing the field of automated construction monitoring.