OBJECT TRACKING ALGORITHM USING PARTICLE FILTER AND FUSION IN MULTI-CAMERA DATASETS

Abstract

The application of object tracking has witnessed explosive increase in tracking techniques and algorithms for different sensors, cameras etc. The data which we obtain from these sources is considerably large. But some of these sources and data are less important .By picking out the most credible sources and efficiently merging data from different data sources like cameras in our case, we can eliminate redundancy issues and obtain accuracy and efficiency in object tracking also data .To deal with this we propose two confidence degrees-Internal confidence & External confidence for prediction of credibility level in every camera. We constructed a camera selection approach wherein only reliable and worthy cameras are selected. Also we implemented a novel data fusion method technique by translating actual data of every frame in form of triangular fuzzy number and uncertainty metrics like information entropy for the fused process is calculated. The result observed is that the proposed theory and algorithm and fusion report is effective and efficient for multiple camera dataset. This study introduces a new method to extract reliable and credible camera data from multiple cameras.Using the internal confidence and external confidence we can determine degree of credibility of each camera. Then using fuzzy set logics we fuse the data from multiple cameras to obtain an efficient data table. Then through uncertainty measures like entropy, we deduce that fused data is more reliable. We have implemented this application on object tracking using three cameras at different angles capturing the same scenario to get a better view dataset. Then the information set from the four camera is fused to get a better overall output.