A multi-feature fusion moving target recognition method based on believability regression reasoning

Features of dynamic target under the complex natural background change drastically, and methods based on single feature recognition could not adapt to the drastic changes, while the method based on multi-feature fusion recognition is one of the important research directions. However, the target distance, scale and background environment vary widely in the process of dynamic target tracking; the basic reliability of multifeature classifiers based on fusion reasoning is unpredictable. This paper proposes a multi-feature fusion recognition algorithm for dynamic target based on reliability regression reasoning. To begin with, target multi-dimensional independent features were extracted; what's more, the basic probability distribution of D-S reasoning based on SVM classifiers was designed according to the mixed matrix distance measure and SVM recognition rate of each feature classifier; furthermore, the relationship between the basic probability distribution and target distance, founded by least square fitting and reliability regression model of D-S reasoning, was acquired. Finally, the method based on multifeature fusion recognition for moving target was fulfilled under the condition of target distance continuous variation at complex environment. Comparative experiments showed that the algorithm has good generalization ability as well as higher efficiency, and the uncertainty of target recognition in the process of dynamic target tracking was reduced to a large extent.