Wall-thickness eccentricity is a major dimensional deviation problem in seamless steel tube production. Although eccentricity is mainly caused by abnormal process conditions in the cross-roll piercing mill, most seamless tube plants lack the monitoring at the hot piercing stage but only inspect the quality of finished tubes using ultrasonic testing (UT) at the end of all manufacturing processes. This paper develops an online monitoring technique to detect abnormal conditions in the cross-roll piercing mill. Based on an image-sensing technique, process operation condition can be extracted from the vibration signals. Optimal frequency features that are sensitive to tube wall-thickness variation are then selected through the formulation and solution of a set-covering optimization problem. Hotelling T2 control charts are constructed using the selected features for online monitoring. The developed monitoring technique enables early detection of eccentricity problems at the hot piercing stage, which can facilitate timely adjustment and defect prevention. The monitoring technique developed in this paper is generic and can be widely applied to the hot piercing process of various products. This paper also provides a general framework for effectively analyzing image-based sensing data and establishing the linkage between product quality information and process information.