Affiliations: [a] CIS Department, Temple University, Philadelphia, PA 19122, USA. E-mail: [email protected], [email protected] | [b] CIS Department and Applied Mathematics Research Center, Delaware State University, Dover, DE 19901, USA. E-mail: [email protected] | Intelligent Systems Division, National Institute of Standards and Technology (NIST), Gaithersburg, MD 20899-8230, USA. Fax: +1 301 990 9688; E-mail: [email protected]
Correspondence:
[*]
Corresponding author.
Abstract: Although a tremendous effort has been made to perform a reliable analysis of images and videos in the past fifty years, the reality is that one cannot rely 100% on the analysis results. With exception of applications in controlled environments (e.g., machine vision application), one has to deal with an open world, which means that content of images may significantly change, and it seems impossible to predict all possible changes. Relying on content-based video analysis may lead to bogus results, since the observed changes may be consequence of unreliable features, and not necessarily of observed events of interest. Our main strategy is to estimate the feature properties when the features are reliable computed, so that any set of features that does not have these properties is detected as being unreliable. This way we do not perform any direct content analysis, but instead perform unsupervised analysis of feature properties that are related to the reliability. The solution pursuit in this paper is to monitor the reliability of the computed features using temporal changes and statistical properties of feature value distributions. Results on benchmark real-life videos demonstrate the capability of the proposed techniques to successfully eliminate problems due to change in light conditions, transition/compression artifacts and unwanted camera motions.
Keywords: motion detection, feature reliability, real time performance evaluation, texture representation, incremental PCA
