Research On The Applications Of Data Fusion In Automobile Recognition

Data fusion technology, which belongs to transverse disciplines, provides a way toprocess data from multiple data sources (signals). It is not a brand new technology. Itsestablishment and development combines a variety of technologies and theories that are wellstudied and developed, such as information theory, probability theory, pattern recognition,neural networks, and so on.In the field of multi-source data processing, a lot of research work has been focused ondata fusion technology in recent years. By data fusion, data from multiple sensors can beprocessed fast, and the ability in target classification and identification can be improved aswell. The work of this thesis has mainly focused on the following areas:(1) The model of data fusion has been studied. Applications of data fusion technology inpattern recognition have been discussed and the feature level and decision level integrationprocesses and algorithms have been described.(2) The scheme that applies data fusion to both feature level and decision level has beenproposed. In this multi-level data fusion approach, the RBF neural network is applied to thefeature level and the fuzzy reasoning algorithm is applied to the decision level in the datafusion process.(3) The proposed scheme is applied to the identification of the type of cars. First in thefeature level, the eigenvalues obtained from collected images are calculated and the RBFneural network is used in data fusion. Then in the decision level, the fuzzy reasoningalgorithm is applied to the RBF neural network outputs and also data from the recordingdevice and speedometer. Finally the decision about the vehicle identification is outputted. Thesimulation results show that the proposed multi-level (based on both the feature level and thedecision level) data fusion scheme can achieve better success rate for identification comparedwith the single-level (based on either the feature level or the decision level) algorithm.