Electronic Image Stabilization is a method to digitally stabilize a video feed to remove unwanted motion by transforming the image. The aim is to create a better viewing experience, reduce bitrate, and simplify further video analytics. This thesis explores different algorithms to stabilize Pan-Tilt-Zoom surveillance cameras made by Axis Communications. Many scenarios are investigated, but the cases when the camera is moving and during high zoom are especially emphasized, as they are the most demanding conditions for image stabilization.

Three alternative real-time stabilization methods are proposed and investigated: one sensor-based, one image-based, and a hybrid between the two. A dataset is recorded at a remote location to evaluate the stabilization algorithms. To compare performance, an evaluation method is developed using motor encoders. The results indicate that the hybrid algorithm works best in most cases except where large foreground objects are moving in the frame. In this case, the hybrid solution tends to shift the image to follow the foreground object, and the sensor-based solution works better with lower drift. During high shaking, the hybrid method performs better regardless.