Compressive-based Video Object Tracking

In computer vision,video object tracking is one of the most core values and the hottest researchtopics of research.Video target tracking applications are challenging,not only in science field butalso in engineering field.Video target tracking in people’s lives has been widely used,such as:intelligent monitoring systems,intelligent transportation,human-computer interaction and so on. Areal-time,very robust tracking algorithm is tried to looked for by using the theory of compressedsensing. The main research contents of this paper are as follows:(1) Atracking algorithm based on compressed sensing is studied in this paper. The Features ofimage are sampled by a measurement matrix. The dimension of the features are selected bysampled is much lower than before.By this method we should deal with features which wassampled from the original features such as avoiding the cost of time on dealing with high dimensionfeatures.(2) A tracking algorithm based on sample is represented by multi-feature. When background ischanged, only one feature can not be well adapted to transform the scene, it will cause the trackerto track the effect of instability.the sample of gray and texture features is used in this paper torepresent sample that to void instability is caused to use only single feature. Two features arebalanced to ensure the stability of the tracker.(3) A tracking algorithm based on weighted classifier based on the feature.A feature weightingapproach to classifier design. Because the impact on the classification of each feature is different, ifjust simply accumulate the classification results directly increases the classification error.For thiscase, depending on the characteristics of the classification results a weighted value, tectonic featureweighting classifier, the cumulative effect of making the smallest error.