適用 3-D 互動人機圖形介面系統的雙核心 Java 處理器設計

Abstract

由於Java語言已經成為目前嵌入式系統上主要應用程式使用的語言,也由於傳統執行Java程式的方式過於複雜,我們希望能夠降低執行程式的時後造成的負擔以及複雜度,來大大提升Java應用程式的效率,因此我們自行開發以一個Java處理器為基礎的SoC平臺,在我們的平台上透過整合Java處理器和一個一般的處理器的異質雙核心平臺,這個平臺能夠幫助我們有效率執行Java程式,這次論文主要時做動態類別載入的機制到平臺上,另外一個時坐的主題是希望能夠整合3-D人機互動介面到平臺上,為了達到虛擬觸控的目的,在論文中特實做相機校正已達到這個目的。

Java is becoming a more popular language for embedded systems. Modem embedded systems do not only execute single task, but often execute many interactive multimedia applications simultaneously. For example, complex UIs based on touch screen are very popular today. The goal of this thesis is to design a heterogeneous dual-core Java platform that can support complex 3-D virtual man-machine interaction. The Java platform is derived from previously published work. In this thesis, hardware and software components are added into existing platform to enable dynamic class loading so that the proposed dual-core Java platform can support large-scale Java programs. In addition, a camera calibration process is also developed in this thesis so that a pair of stereo cameras can be used to capture the 3-D motion of the human operator. With 3-D motion input capability, the proposed platform will be capable of efficient execution of 3-D interactive GUI systems. The proposed dynamic class loading mechanism has been implemented on the Xilinx Virtex-5 ML507 FPGA development board, and verified the proposed design with a subset of Embedded Caffeine Mark. The performance of our design (without data cache) is about 3~ 9 times faster than CVM running on RISC processor with data cache at the same clock rate. We also verified the camera calibration algorithm using simulated image sequences and the errors of camera calibration process is about 5~10 mm for a virtual object located about 700 mm away from the camera.