Abstract This paper presents an optimized feature tracking mechanism. The proposed mechanism outperforms our previously proposed algorithms which are on the basis of feature matching. Specifically, for the work process of the proposal, extracting the FAST feature from input images at first. Then, by applying optical flow algorithm, the feature points appeared in different images will be tracked and marked out in the images. Afterwards, we apply the optimized PnP algorithm to estimate the camera motion. The results of the evaluation experiments which are conducted on raspberry pi indicate that our proposed mechanism provides a 16.7% improvement in time-efficiency compared with our previous work. Furthermore, the proposed mechanism still keeps the estimation error at a lowest level, as our previous work did. In summary, the proposed mechanism achieves higher time-efficiency and good estimation accuracy on IoT devices, which only have constrained sources.