Camera Movement Cancellation In Video Using Phase Congruency and An Fft-Based Technique

One of the interesting fields of video processing is motion detection and human action detection (HAR) in video. In security systems or smart city monitoring where both objects and the camera may be moving, removing camera movement is very important to increase accuracy in extracting motion features. HAR sys-tems usually use image matching and image registration algorithms to remove camera movement in a video. In these methods, source (fixed) picture is com-pared with moved picture, and the best match is determined geometrically. Some assumptions, such as ignoring translation, rotation or scale, are used to increase speed. In video processing, due to the existence of a set of frames, we can correct errors using previous data, but at the same time, we need a fast frame registration algorithm. According to the above explanations, this research proposes a method to detect and minimize camera movement in video using phase information. In addition to having the acceptable speed and the ability to be implemented online, the pro-posed method, by combining texture and phase congruency, can significantly in-crease the accuracy of detecting the objects in the scene. The proposed method was implemented on a HAR database, which includes camera movement, and its ability to compensate for camera motion and preserve object motion was verified. Finally, the speed and accuracy of the proposed method are compared with a number of the latest image registration methods and its efficiency in the camera movement cancellation and in the field of execution time is shown.