HMM 使用 Accord.Net 进行手势识别答案

【问题标题】：HMM using Accord.Net for gesture recognitionHMM 使用 Accord.Net 进行手势识别
【发布时间】：2014-02-05 17:09:30
【问题描述】：

我正在从事手势识别项目。我的数据集由 4 个不同的手势组成，每个手势集包含大约 70 张图像。我为每张图像提取了 4 个特征。我正在尝试使用 Accord.Net 来实现 HMM，并且我知道我将需要 4 个 HMM，每个手势一个，但我不确定如何构建用于学习/训练的特征向量序列阶段。有人知道怎么解决吗？

这是序列的简单代码：

double[][] sequences = new double[][] 
{
    new double[] { 0,1,2,3,4 }, // This is the first  sequence with label = 0 
    new double[] { 4,3,2,1,0 }, // This is the second sequence with label = 1
};

// Labels for the sequences 
int[] labels = { 0, 1 };

【问题讨论】：

标签： c# gesture-recognition hidden-markov-models accord.net

【解决方案1】：

您是对的，每个手势都需要一个 HMM。但是，如果您使用 HiddenMarkovClassifier 类（它是在您尝试检测的每个类之后创建的多个 HMM 的包装器），框架已经可以为您提供这种构造。

如果每张图像有 4 个特征，则需要假设能够对多变量特征进行建模的概率分布。一个简单的选择是假设您的特征彼此独立，并且它们中的每一个都遵循正态分布。

因此，您可以使用以下示例代码来创建模型。它假设您的数据库只有两个训练序列，但实际上您必须有很多个。

double[][][] sequences = new double[][][]
{
    new double[][] // This is the first  sequence with label = 0
    { 
        new double[] { 0, 1, 2, 1 }, // <-- this is the 4-features feature vector for
        new double[] { 1, 2, 5, 2 }, //       the first image of the first sequence
        new double[] { 2, 3, 2, 5 },
        new double[] { 3, 4, 1, 1 },
        new double[] { 4, 5, 2, 2 },
    }, 

    new double[][] // This is the second sequence with label = 1
    {
        new double[] { 4,  3, 4, 1 }, // <-- this is the 4-features feature vector for
        new double[] { 3,  2, 2, 2 }, //       the first image of the second sequence
        new double[] { 2,  1, 1, 1 },
        new double[] { 1,  0, 2, 2 },
        new double[] { 0, -1, 1, 2 },
    }
};

// Labels for the sequences
int[] labels = { 0, 1 };

上面的代码显示了如何设置学习数据库。现在，一旦设置好了，您就可以为 4 个正态特征（假设正态分布之间的独立性）创建一个隐藏马尔可夫分类器

// Create one base Normal distribution to be replicated accross the states
var initialDensity = new MultivariateNormalDistribution(4); // we have 4 features

// Creates a sequence classifier containing 2 hidden Markov Models with 2 states
// and an underlying multivariate mixture of Normal distributions as density.
var classifier = new HiddenMarkovClassifier<MultivariateNormalDistribution>(
    classes: 2, topology: new Forward(2), initial: initialDensity);

// Configure the learning algorithms to train the sequence classifier
var teacher = new HiddenMarkovClassifierLearning<MultivariateNormalDistribution>(
    classifier,

    // Train each model until the log-likelihood changes less than 0.0001
    modelIndex => new BaumWelchLearning<MultivariateNormalDistribution>(
        classifier.Models[modelIndex])
    {
        Tolerance = 0.0001,
        Iterations = 0,

        FittingOptions = new NormalOptions()
        {
            Diagonal = true,      // only diagonal covariance matrices
            Regularization = 1e-5 // avoid non-positive definite errors
        }

        // PS: Setting diagonal = true means the features will be
        // assumed independent of each other. This can also be
        // achieved by using an Independent<NormalDistribution>
        // instead of a diagonal multivariate Normal distribution
    }
);

最后，我们可以训练模型并在学习数据上测试其输出：

// Train the sequence classifier using the algorithm
double logLikelihood = teacher.Run(sequences, labels);


// Calculate the probability that the given
//  sequences originated from the model
double likelihood, likelihood2;

// Try to classify the 1st sequence (output should be 0)
int c1 = classifier.Compute(sequences[0], out likelihood);

// Try to classify the 2nd sequence (output should be 1)
int c2 = classifier.Compute(sequences[1], out likelihood2);

【讨论】：