High Performance and Novel Target Recognition Algorithm in the Sparse Measurement Space

This Phase II project focuses on the development of a compact, light-weight, low cost, low power, high resolution, high frame rate, low motion blur, and high dynamic range (HDR) imager for target recognition. We have both innovative hardware and software algorithms. In the hardware, we propose an advanced compressive sensing CMOS imager that has several key features. First, the imager has pixel-wise exposure control and because of this, the imager is low power, light weight, compact, and has high frame rate, high dynamic range, and low motion blur. Second, the imager is also scalable to high resolution. We will build several variants of this imager, including a high resolution grey level imager, a high dynamic range (HDR) imager, and a high resolution color imager. In the software, we propose to perform target recognition directly in the compressive sensing domain. This will allow faster and more accurate recognition performance than conventional methods. Target/pattern detection, classification, and recognition applications will benefit more by incorporating such class-specific discriminative information than merely by conventional sparse signal recovery followed by a conventional classification strategy. Hence, we focus on maximizing the discriminability by enforcing meaningful adaptive class-specific priors/constraints directly in the data measurement domain.