Object tracking is an important technology of image and video processing for many applications, including vision-guided automation, automatic target identification, object-based video compression, and face recognition. A fundamental tool that underlies the technology of object tracking is machine classification. Recent developments of binary psi-learning allow us to further achieve higher generalization accuracy for nonseparable and multiclass cases. This proposal presents an interdisciplinary research plan to address the problem of multiple object-tracking via this new learning tool. The project will investigate how the accuracy of psi-learning can be maximized by studying its generalization ability as well as methods for assessment. Specific desired outcomes of the project are (a) creation of a stable foundation for psi-learning and further development of learning theory, optimization theory, and algorithms, and (b) specific development of mechanisms for object tracking and extraction in multimedia compression.
The proposed project is expected to have broad impacts to education, research, economy, and society at large. In particular, the technology developed in this project is widely applicable to scientific and engineering frontiers, including face-recognition, target identification, and cancer genomics classification. Plans for technology transfer are proposed to benefit the economy. The proposed educational program will train students in an interdisciplinary area of statistics and electrical engineering. Success of this project will bring tremendous benefits to fundamental scientific research, high-performance computing, and information technology, and have significant broad impacts to society at large.