Evidential Fusion for Gender Profiling

Gender profiling is a fundamental task that helps CCTV systems to<br/>provide better service for intelligent surveillance. Since subjects being detected<br/>by CCTVs are not always cooperative, a few profiling algorithms are proposed<br/>to deal with situations when faces of subjects are not available, among which<br/>the most common approach is to analyze subjects’ body shape information. In<br/>addition, there are some drawbacks for normal profiling algorithms considered<br/>in real applications. First, the profiling result is always uncertain. Second, for a<br/>time-lasting gender profiling algorithm, the result is not stable. The degree of<br/>certainty usually varies, sometimes even to the extent that a male is classified<br/>as a female, and vice versa. These facets are studied in a recent paper [16] using<br/>Dempster-Shafer theory. In particular, Denoeux’s cautious rule is applied for<br/>fusion mass functions through time lines. However, this paper points out that if<br/>severe mis-classification is happened at the beginning of the time line, the result<br/>of applying Denoeux’s rule could be disastrous. To remedy this weakness,<br/>in this paper, we propose two generalizations to the DS approach proposed in<br/>[16] that incorporates time-window and time-attenuation, respectively, in applying<br/>Denoeux’s rule along with time lines, for which the DS approach is a special<br/>case. Experiments show that these two generalizations do provide better results<br/>than their predecessor when mis-classifications happen.