| Abstract | Due to limited resolution and quality of surveillance video, tracking and recognizing of objects in surveillance video requires techniques for accumulation of information about the object over time. The simplest of these techniques is histograms, which computes the distribution of pixel values over time and which is frequently used for tracking uniformly coloured objects such as faces. Another well-known technique for the purpose is correlograms, which learns pixel values and their spatial relationship to yield better discriminative power. However, this technique also lacks a comprehensive learning ability, because it updates the information about the object, expressed in terms of cooccurance matrices, using the currently observed pixels only and ignoring the preceding learning history. Associative neural network based memorization can thus be seen as a more sophisticated mechanism for accumulation of information about an object in a video sequence. This presentation describes how to use the Open Source Associative Neural Network code for tracking and recognition of objects in video. Two demos showing multiple-face tracking and classification in low-resolution video are shown. |
|---|
