Adaptive Fuzzy Weighted Template Matching Using Invariant Features for a Tracking Application

Abstract

One of the common research topics for surveillance systems is tracking of a target object. Motions of both observer and target make the problem difficult to handle. In this paper, proposed algorithm tracks a target object on consecutive frames taken from a camera embedded on an unmanned aerial vehicle. Adaptive-fuzzy weighted sum of square distances (SSDs) are utilized for template matching process. SSD is computed by the pseudo-Zernike moments of the reference template and possible matching template around the search point. Fuzzy weights are obtained by taking the advantage of circular projection features. An adaptive fuzzy system is realized by updating the weights of the defuzzifier. Proposed method provides center point of the best matching template which represents the target object to be tracked. Tracking is accomplished under the presence of illumination changes and parallax, and it is seen that the proposed algorithm yields satisfactorily small errors.

A novel unmanned aerial surveillance scheme

“…For a comparison, NCC algorithm [13,14] and two stage template matching (TS-TM) algorithm discussed in [15] are applied on the same test video with same template size and search area size. The work in [16] improves TS-TM by proposing an adaptive fuzzy weighted template matching (AF) algorithm with improved feature space. The experimental results of [16] were demonstrated with three different parameter settings in fuzzy module so we also included AF comparison with all presented settings.…”

“…The work in [16] improves TS-TM by proposing an adaptive fuzzy weighted template matching (AF) algorithm with improved feature space. The experimental results of [16] were demonstrated with three different parameter settings in fuzzy module so we also included AF comparison with all presented settings. In NCC, TS-TM, AF, and proposed algorithms, next search point is determined by the previous selected point of the best matching template.…”

“…Hence, as shown in ( 15), the weight corresponding to the closest previous frame is larger than the others, meaning that its effect is more than the effects of others on choosing the next search point. Based on (15), the search point in the following frame can be expressed as in (16).…”