The practical deployment of pervasive health monitoring requires a close integration of body sensor networks (BSNs) with intelligent ambient sensing. To this end, vision sensors with low cost and minimal power consumption provide an attractive means of activity tracking and capturing early signs of disease progression through changes in gait and posture. The purpose of this paper is to present an imagebased modeling technique for generating a subject-specific simulation environment that allows systematic development of novel vision algorithms that can be implemented by low power video sensors with distributed processing and known ground-truth.