透過您的圖書館登入
IP:18.223.24.92
  • 學位論文

基於區域分割實現即時仿人視覺系統之研究

Real-time Humanoid Visual System based on Region Segmentation for obstacle determination

指導教授 : 施文彬

摘要


本論文建構出一套完整的即時視覺系統,配合自行研製之仿人眼機構的特性發展適合的控制方法。該人眼機構具有五個自由度以控制雙眼的水平與垂直轉動乃至眼皮的開闔。使用壓克力做為材料以方便電腦輔助製造並達到輕量化的要求,採用的馬達以及控制電路也都選用了輕型微小的產品。 首先利用小型的網路攝影機晶片輸入影像,之後視覺軟體演算法的部份以區域為單位進行不同視角影像間的配對,並藉由找到對應區域的視差來推算出該區域代表物體與仿人眼機構之間的三維距離。為了在取得良好分割效果的同時仍具即時運算的能力,本研究採取k-mean分類演算法為基底作色彩分類,並改良色彩空間使色差的計算均勻化來改善其效果。區域擷取的部分也設計了加速的機制使得實際搜尋各區域範圍的過程僅需掃描各區域邊界點一次。 區域配對的結果將化成對應的形心配對作為距離估計的基礎,其計算結果的誤差會受到初始影像感測器的參數測定以及仿人眼機構轉動角度的控制精度所影響。為了增進精度該機構配置上了角度感測計以及小型的類比轉數位板作即時水平旋轉角度回饋,對於垂直旋轉部份則因難以裝配硬體角度計,而另外採取特殊的軟體修正方式,並同時活用了座標轉換推導下所產生之額外的等式。 為了最終的運行速度能達到即時的要求,本研究加入了以Nvidia顯示卡以及對應的CUDA開發工具,將原本的程式適當的平行化來運用GPU的高效運算能力。最終的視覺系統能在一般的個人電腦平台上達到30Hz以上的運作週期,滿足一般認知上的嚴格即時運作定義。

並列摘要


This thesis proposes one real-time visual system completely, and the suitable control methods are developed for the especial humanoid eye mechanism by our self. This eye mechanism has five degree of freedoms to control the horizontal and vertical rotations of two eyeballs and opening of two eyelids. PMMA is chosen as the material for convenience of CNC machining and its lightness. Those used motors and control boards are also mini and light. The analyzed images are inputted from mini webcams, and then the visual program use region-based corresponding result between images with different viewpoints and disparities of those corresponded regions to calculate the spatial distances between the eye mechanism and those objects represented by corresponded region pairs. For the balance between efficacy of region separation and computing efficiency to satisfy real-time operation, this research uses the algorithm of k-mean classification as the basis to classify image pixels by color statistics and also improves the color space to let it more uniform in description of chromatism for better efficacy. The accelerative program mechanism for regions extraction is also designed to let the passing times of each region boundary decrease to only one in actual searching stage. Those centroid pairs from corresponded regions are considered that the basis of distance determination. The errors of calculated distance are affected by parameters calibration of image sensors and precision of eyeballs’ rotational angles control. Therefore, the eye mechanism is also installed with angle gauges and analog-to-digital board to achieve real-time feedback of horizontal rotations. The revision of vertical rotation control by software is designed because the difficulty of installation and applying of the extra equation from coordinate transformation. To meet the requirement of real-time operation finally, this research also involves the help by Nvidia graphic cards and CUDA toolkits. After suitable parallelization of original serial program and operation on GPU of graphic cards, the efficiency of visual system obtains 5 to 10 times improvement by the higher ability in general computation of current GPU based system. Finally, the operation frequency of this visual system achieves more than 30 Hz which meets the general definition of real-time operation.

參考文獻


[1] J. Fan, D. K. Y. Yau, A. K. Elmagarmid, and W. G. Aref, “Automatic Image Segmentation by Integrating Color-Edge Extraction and Seeded Region Growing,” IEEE Trans. Image Process., vol. 10, no. 10, pp. 1454-1466, 2001.
[2] N. Otsu, “A Threshold Selection Method from Gray-Level Histograms,” IEEE Trans. Syst. Man Cybern., vol. 9, no. 1, pp. 62-66, 1979.
[3] J. Kittler and J. Illingworth, “Minimum Error Thresholding,” Pattern Recognit., vol. 19, no. 1, pp. 41-47, 1986.
[4] M. Sharifi, M. Fathy, and M. T. Mahmoudi, “A Classified and Comparative Study of Edge Detection Algorithms,” Proc. ITCC 2002, U.S.A, pp. 117-120, April 2002.
[5] J. Canny, “A computational approach to edge detection,” IEEE Trans. Pattern Anal. Mach. Intell., vol. PAMI-8, No. 6, pp. 679-698, 1986.

延伸閱讀