Disaster mitigation is a significant issue where modern wireless network systems are expected to play a crucial role. It is believed that Visual IoT is one of the key techniques because of its monitoring abilities of urban and rural areas. The outdoor Visual IoT systems generally transfer a large number of footage every day. To detect information from large-size footage datasets, automatic time subtractions between frames are effective. However, to extract even tiny difference between frames, camera shake causes a serious damage. In this study we first survey long-durational footage transmitted from outdoor cameras installed in a city to examine that the stabilization techniques based on feature keypoints are effective to camera shake. Based on this survey we define a matching index to judge if the stabilization is of use or not for every couple of frames. The index is implemented with help of a camera calibration library in OpenCV using AKAZE feature. We then propose a method to stabilize footage continuously obtained by outdoor Visual IoT systems. Using the matching index, we examine one-day footage to find that stabilization is occasionally not applicable in case when the matching index is relatively larger or smaller than 1 (when matching is complete) or the number of matching pairs using AKAZE keypoints are too few. According to the results we set a threshold value of the matching index. We finally perform this technique to footage recorded on a couple of strongly windy days. The efficiency is numerically and visually confirmed on each footage successfully.