A Single-class Support Vector Machine Translation Algorithm To Compensate For Non-stationary Data In Heterogeneous Vision-based Sensor Networks

This paper develops a translation algorithm that adapts an existing support vector machine (SVM) to observations that have a different probability distribution than originally trained with. The primary advantage of this algorithm is that the re-training can be avoided. The support vector translation algorithm can be used in a fully distributed vision-based sensor network for target classification and tracking. Preliminary results are discussed and planned future work is briefly outlined.