Determination of localised motion in live video data
With CVB Optical Flow developers can benefit from an efficient tool for the determination and analysis of movement within an image sequence. The tool provides the detection, data extraction and display of local motion vectors with an accuracy of less than 2° mean deviation when determining the direction and less than 1/10 pixel deviation when determining the distance moved.
Determination of localised movement in video is of major importance for viewing and analysing motion in the environment. Computer supported determination and analysis of moving components in images, often called "optical flow" is a highly complex task and it is challenging to undertake full analysis at real time frame rates. For this reason it is only rarely available in imaging libraries.
Compared with conventional methods for detecting movement, such as calculating the difference between images, CVB Optical Flow not only allows detection, but also provides the data allowing for quantitative analysis of movement of image content. Featuring robust algorithms it is also extremely tolerant against poor and rapidly changing light conditions, as well as vibration. The tool is ideally suited not only for industrial applications such as bulk material flow and process monitoring, but also in traffic or security and surveillance applications to determine traffic flow, illegal movement and potential hazard detection.
The algorithm is based on a specially adapted version of the so-called block matching technique. Using this method the image is divided into blocks within which the displacement of the image content is determined. The flow vectors are calculated for each fixed grid point in the image. Blocks can overlap arbitrarily to reach higher spatial resolutions while the block size as well as step size are freely selectable.