Matching of video objects taken from different camera views by using multi-feature fusion and evolutionary learning methods

摘要

Now a day the identifying of an object by using more than one camera with different views is very necessary. The images or video capture by these cameras should be non-overlapped. Therefore, to capture the specific object in the wide area video surveillance it is very difficult task. Most of the information is lost due to the different cameras and movement of the object. Generally the visual information is to get from the cameras. This paper presents a novel algorithm which is used to obtain the different features of an object and compare these features to the appearance features of individual objects. Firstly, the representation of a Competitive Major Feature Histogram Fusion (CMFH) is used to find the appearance of model which helps to characterize the potentially of the matching objects. In front of different cameras, the appearances of an object can change over time to time and hence it is necessary to continuous update these models. For this purpose in this paper modified Incremental General Multi Category Support Vector Machine (IGMSVM) algorithm is present. The IGMSVM algorithm updates the models online and matches the online objects with the objects obtained from camera by using the classification method. To calculate accurate classification of a model, small amount of samples of an object are required. In the experimental results there are different tests are performed on different database where an objects images are taken at different view point, illumination and the poses. The proposed algorithm is capable to obtain correct match of typical object.