Hopfield模式之基因演算法在獲取物體三維位置之應用

Positioning 3D Objects Using Genetic Algorithm with Hopfield Neural Network

劉孝忠(Hsiao-Chung Liu)

《中華科技大學學報》 60期 (2014/10) Pp. 1-24

摘要

本文探討空間中物體3D位置的析出，在3D位置的析出問題方面，採用一個簡單的三角量測方法，再配合立體取像法的原理結合類神經網路之特性，可以很容易的求得一個物體的3D位置。其過程如下，先利用相機擷取左右兩張影像，轉換成數位訊號，再經過高斯拉布拉斯運算子分別求出左右兩張影像之零交點，經由影像梯度差篩選，除去不必要的雜訊。再根據零交點連續的原理，將零交點分成16種型別。在作左右影像特徵點比對時，利用Hopfield模式之基因演算法最佳化類神經網路。左右影像特徵點比對之問題可視為一組合最佳化（combinatorial optimization）過程，先根據比對之限制及要求定義一個能量函數，此能量函數需符合三個條件，平滑性（比對成功點之鄰域像差不可變化太大）唯一性（物體上任何一點只能有一個比對點）、相似性（兩個相互比對成功之特徵點須具有相似之零交點特性）。然後將此能量函數對應至一個二維之類神經網路，以0或1來代表網路中神經元是否比對成功，網路中之神經鍵值隱含著比對的限制和要求，利用能量函數之遞減來疊代神經元之值，當此網路達到平衡，也就是此能量函數達到極小值，即可得到最佳之比對結果。為了精進比對成功的百分比，在此利用臨域像差篩選，以篩選出比對錯誤之特徵點，接著利用水平型零交點像差之內插法完成所有特徵點像差之求取。最後再應用一個三角量測之原理即可得到所需量測物體之特徵點的3D位置。

關鍵字

Hopfield類神經網路 ； 基因演算法

並列摘要

An algorithm is presented herein to obtain the 3D position of an object. The proposed algorithm can obtain an object 3D position by using the Genetic Algorithm with a Hopfield neural model. Correspondence problem is the primary concern of stereo vision. A combinatorial optimization approach is used to resolve the correspondence problem for a set of features extracted from a pair of stereo vision. An energy function is defined to represent the three solution's constraints: (1) smoothness, disparity values change smoothly, (2) uniqueness , each point in an image should be assigned at most one disparity value, and (3) similarity, pair of matched zero-crossing points must have the same feature; the energy function is then mapped onto a 2D neural network. Each neuron in the network represents a possible correlation between a feature in the left image and another one in the right image. The features are zero-crossing points that are extracted using the LOG operator. Zero-crossing points are classified into sixteen patterns according to their local connectivity. The difference value of the sign and direction between a matched pair of zero-crossing can be used to setup the neural node's iteration rule. The network is assumed to be at its stable state when no change occurs in the neurons' state. Finally, the threshold of the neighbor's disparity(NDT) is used to enhance the precision of the corresponding situation. Once all the correspondence points are found, obtaining the 3D object position is a simple matter of triangulation.

並列關鍵字

Hopfield neural model ； Genetic algorithm