Research of a Framework for Flow Objects Detection and Tracking in Video

摘要

The flow objects are ubiquitous in nature, and the detection and tracking of flow objects is very important in the field of machine vision and public safety, so building a framework for the detection and tracking is more advantageous for this research. For this demand, a systematic framework is proposed. First, the foreground can be detected by GMM (gaussian mixture model) and SNP (statistical nonparametric) algorithm, and candidate regions can be determined by static features extracted in the foreground. Second, all these candidate regions should be combined and tracked. At last, dynamic features of the tracked regions should be extracted and whether it is flow objects or not should be confirmed. To solve the problem of combination of adjacent small regions and the multi-objects matching, similar regional growth algorithm and the method for tracking multiple targets are put forward. To verify the effect of the framework, a lot of experiments about smoke, fire, and rain are implemented.