Motion detection using optical flow

Motion detection which is one of the important fields in computer vision has many applications in daily life such as security systems, traffic surveillance. ?Optical Flow?, one of the first proposed approaches used for motion detection, still gives more robust and accurate results then many approaches proposed lately. Moreover, many motion detection methods are based on ?Optical Flow?. The first part of this thesis introduces fundamental concepts within general survey. Then optical flow methods and their structures are introduced. Furthermore, special cases that ?Optical Flow? can not determined accurately are mentioned. Optical flow techniques can be classified into three groups: 1. Differential techniques, 2. Frequency-based techniques and 3. Matching techniques. Second part of this thesis introduces Differential techniques: global differential approach (Horn and Schunk method) and local differential approach (Lucas and Kanade method). Motion estimations which are obtained from MATLAB implementations using real image sequences are shown in vector fields and energy fields. Additionally, Gaussian Pyramid is discussed and with using Gaussian Pyramidal implementation of Lucas and Kanade method, more robust and accurate results are obtained. Keywords: Optical flow, differential techniques, motion detection.