Motion Analysis and Tracking
The measurement of object or camera motion from video sequences is an important component in many applications. For example, in computer vision systems it enables the identification and tracking of the objects that make up a scene; while in video data compression it provides a means of reducing redundancy - knowing the motion of an object allows its position in successive frames to be predicted, removing the need to retransmit identical frame data and leading to a reduction in the bit rate required to transmit the video. Other applications include the generation of high resolution and panoramic images from video and the automated building of virtual reality environments.Recovering motion from video sequences is a complex task. Factors such as the ambiguity resulting from 3-D to 2-D projection, poor colour contrast, and low spatial-temporal resolution, mean that sophisticated techniques are required in order to obtain reliable motion information. We are working on a range of approaches to tackle such problems. These include:
- the use of affine models to estimate and track complex 2-D motions, such as rotation, dilation and shear.
- techniques to deal with multiple component motion regions, such as those encountered at boundaries or within transparent objects.
The techniques being developed are suitable for any application involving motion estimation, although our primary concern is with their use in video data compression. This work aims to enhance existing approaches and to form the basis of the next generation of compression algorithms, particularly those employing object and model-based techniques.