Integrating social robotics into the construction industry, particularly in the context of Industry 5.0, faces several challenges in creating complex environments that seamlessly blend human and machine interactions. In this regard, the emergence of intelligent and expert systems holds promising technologies to enhance construction tasks focused on robots and workers in 3D printing applications. This work compares several methods of convolutional neural network-based object detectors designed to identify distinct construction assets and workers within the dynamic environment of 3D printing. To this end, different versions of the You Only Look Once v8 (YOLO v8) algorithm have been implemented, trained, and experimentally tested using several images captured within dynamic construction environments. Furthermore, we present an in-depth comparison between YOLO v8 and its preceding versions, namely YOLO v7 and YOLO v5. Experimental results disclosed the high performance of the proposed approach in effectively detecting three distinct entities (workers, robotic platforms, and building elements), achieving a precision rate of up to 98.8%.