Fast tracking algorithm using modified potential function

This study introduces a new potential function-based modelling approach for real-time object tracking with single camera. Real-time tracking requires the least complex techniques for processing and classification and still provide accurate results. Particle filter-based algorithms allow accurate estimations of the displacement and scaling of the object for tracking, but at the cost of high computational complexity and complicated modelling. Also, the existing single-camera tracking systems lack the ability to predict the direction of motion of the object and their performance is significantly affected by occlusions. This study proposes a new method to address these four key issues. The method is principally based upon the potential function, which has been modified for motion image sequences. Potential function uses the current estimates of non-linear scaling and drift vector with a priori knowledge of the object to compute the tracking parameters in the form of diffusion matrices. The concept of attractors and repellers inside a potential field has been used in analogy to classify different directions of motion in the image plane, such that the object tends to drift towards the attractors and away from repellers. Attractor for every consecutive pair of frames is estimated using the set of transformations (displacement and scaling) occurred due to the motion in a particular direction. The proposed technique works well with minimal tracking errors and a computational complexity of O(1).