Quantum-classical generative adversarial networks based computer vision solution for human and construction vehicle detection

Detection of human and construction vehicle is an important step when it comes to automation of safety and productivity on construction sites. The technique of Computer Vision in its conventional forms has issues in cases where shadows, noise or large differences in illumination and occlusions occur. Although there are such challenges, the application of classical Generative Adversarial Networks (GANs) has promising solutions to such problems, but the training and optimization may take longer time and may not converge. The arrival of quantum computation allow for improvements on GAN characteristics through dimensional modeling and optimization at a high speed. Nevertheless, fully quantum models are still limited by hardware and can only be solved with the help of mixed-classifications. For construction vehicle detection, this research work suggests a combined Quantum-Classical Generative Adversarial Network (QC-GAN) approach. The connection component introduces quantum aspects into the procedure to capture elaborate features of the image data and enhance feature engineering and its generalisability. To enhance functioning of the QC-GAN, we incorporate the Leopard Seal Optimization (LSO) algorithm to function based on hunting attributes of leopard seal. The present LSO algorithm successfully navigates through the optimization landscape for better training stability and faster convergence. The introduced approach attains higher accuracy as 99%.