There is high demand for heavy equipment in civil infrastructure projects and their performance is a determinant of the successful delivery of site operations. Although manufacturers provide equipment performance handbooks, additional monitoring mechanisms are required to depart from measuring performance on the sole basis of unit cost for moved materials. Vision-based tracking and pose estimation can facilitate site performance monitoring. This research develops several regression-based deep neural networks (DNNs) to monitor equipment with the aim of ensuring safety, productivity, sustainability and quality of equipment operations. Annotated image libraries are used to train and test several backbone architectures. Experimental results reveal the precision of DNNs with depthwise separable convolutions and computational efficiency of DNNs with channel shuffle. This research provides scientific utility by developing a method for equipment pose estimation with the ability to detect anatomical angles and critical keypoints. The practical utility of this study is the provision of potentials to influence current practice of articulated machinery monitoring in projects.