The feasibility of real-time non-invasive spatio-temporal temperature estimation from pulse-echo diagnostic ultrasound data has been previously demonstrated in stationary phantoms. The method is based on first estimating the axial shifts of the RF-echo data due to local changes in the speed of sound and thermal expansion in the propagating medium, and then differentiating these estimates along axial direction to obtain the temperature rise map. In a clinical setup, however, translation, rotation and deformation affect the estimates. In this paper we introduce an algorithm to compensate for tissue translation and uniform deformation along the axial and lateral directions of the ultrasound imaging plane. This is achieved by separating the components of the time-shift map due to temperature rise (a local effect, occurring within the vicinity of the heated region) from the component due to translation and deformation (effect observed over a larger region). A rubber phantom experiment was designed where high intensity focused ultrasound was used to generate localized heating while motion was applied to the phantom and/or imaging transducer. Temperature profiles were successfully estimated while the phantom was translated by 30 mm and axially deformed by 13%.