A target tracking algorithm using a sensor with biological vision features

One of the most difficult problems in a video tracking system is the speed required for real-time operation. In this paper, a fast tracking algorithm to determine displacements of points (pixels) on object surfaces between successive frames, a sequence of image frames representing a time-varying scene is described. The displacement is represented by a four-dimensional vector whose elements are two dimensional translation, rotational angle on the x-y plane, and scaling factor (zoom parameter). The algorithm was developed in conjunction with research on sensors for machine vision with biological vision features. The proposed sensor has two configurations: a uniform rectangular grid and a nonuniform logarithmic spiral grid. The rectangular grid is very well suited to translational motion in two-dimensions, whereas the logarithmic spiral grid has been proven to be appropriate for rotation and scaling. Both configurations are utilized to estimate three-dimensional motion as defined above. The algorithm developed is called a one-dimensional correlation algorithm and has recursive, spatio-temporal, correlation characteristics. It also possesses simplicity and separability properties. These properties make it easy to design and implement it in a highly parallel fashion. A speed comparison analysis shows that this algorithm can save 40% of computation time, or more, over the conventional correlation algorithm if implemented in a general purpose serial processor. Furthermore, the simulation results on a real image demonstrate that this algorithm possesses a good noise immunity property. That is, it can be applied directly to a predefined moving area without segmentation. The properties and capabilities of the algorithm make it especially suited for robotics applications with the camera mounted at the end effector. In such a situation the work space and conveyor belts would usually be horizontal and vertical, rotation and scaling would occur in a plane perpendicular to the optical axis of the sensor.