传说中的BP手势识别

yangff · 发表于 2010-12-31 21:55:30

加入我们，或者，欢迎回来。

您需要登录才可以下载或查看，没有帐号？注册会员

x

根据一个C#代码和一本书里面的翻译过来的
识别率不高……
另外为了提高效率，神经网络是事先Dump出来的
还有只能识别0~9
原始代吗是这个http://blog.linjian.org/articles/bp-mouse-gesture-recognition/
就是这样~
至于怎么用……
复制所有代码
DrawBox.new(x,y,width,height)
然后这只是个简易的版本，不解释……一般使用是够了……更多的功能……自己看着办吧……
NetworkFunctions.initData是构建网络
NetworkFunctions.save是保存网络
NetworkFunctions.load是读取网络
其他的什么常量啊7788的自己看着办被……

module Useful
NUM_PATTERNS=10
NUM_VECTORS=12
MATCH_TOLERANCE=0.96
ACTIVATION_RESPONSE=1.0
BIAS=-1.0
ERROR_THRESHOLD = 0.003
LEARNING_RATE = 0.5
NUM_HIDDEN_NEURONS = 6
WITH_MOMENTUM = false
MOMENTUM = 0.9
WITH_NOISE = false
MAX_NOISE_TO_ADD = 0.1
InitNames=["0","1","2","3","4","5","6","7","8","9"]
InitPatterns=
[
[ -0.96, 0.29, -0.71, 0.71, -0.29, 0.96, 0.29, 0.96, 0.71, 0.71, 0.96, 0.29, 0.96, -0.29, 0.71, -0.71, 0.29, -0.96, -0.29, -0.96, -0.71, -0.71, -0.96, -0.29 ],
[ 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1 ],
[ 1, 0, 1, 0, 1, 0, 1, 0, -0.71, 0.71, -0.71, 0.71, -0.71, 0.71, -0.71, 0.71, 1, 0, 1, 0, 1, 0, 1, 0 ],
[ 1, 0, 1, 0, 0.91, 0.41, 0.41, 0.91, -0.41, 0.91, -0.91, 0.41, 0.91, 0.41, 0.41, 0.91, -0.41, 0.91, -0.91, 0.41, -1, 0, -1, 0 ],
[ -0.32, 0.95, -0.32, 0.95, -0.32, 0.95, -0.32, 0.95, -0.32, 0.95, -0.32, 0.95, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0 ],
[ -1, 0, -1, 0, -1, 0, 0, 1, 0, 1, 1, 0, 0.91, 0.41, 0.41, 0.91, -0.41, 0.91, -0.91, 0.41, -1, 0, -0.91, -0.41 ],
[ -0.95, 0.32, -0.71, 0.71, -0.35, 0.94, 0, 1, 0.35, 0.94, 0.71, 0.71, 0.95, 0.32, 0.91, -0.41, 0.41, -0.91, -0.41, -0.91, -0.91, -0.41, -0.91, 0.41 ],
[ 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1 ],
[ -0.89, -0.45, -0.89, 0.45, 0, 1, 0.89, 0.45, 0.89, 0.45, 0, 1, -0.89, 0.45, -0.89, -0.45, 0, -1, 0.89, -0.45, 0.89, -0.45, 0, -1 ],
[ -1, 0, -1, 0, -1, 0, 0, 1, 0, 1, 1, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 1 ]
]
def self.randomClamped
a=rand(11)/10.0
b=rand(11)/10.0
return a-b
end
module RUN_MODE
LEARNING=1
ACTIVE=2
UNREADY=3
TRAINING=4
end
end
class Neuron
attr_accessor:numInputs
attr_accessor:weights
attr_accessor:prevUpdate
attr_accessor:activation
attr_accessor:error
@numInputs=1
=begin
public Neuron(int _NumInputs)
{
NumInputs = _NumInputs + 1;
Activation = 0;
Error = 0;
Weights = new List<double>();
PrevUpdate = new List<double>();
//生成随机权重
for (int i = 0; i < NumInputs; i++)
{
Weights.Add(Useful.RandomClamped());
PrevUpdate.Add(0.0);
}
}
=end
def initialize(numInputs)
@numInputs=numInputs+1
@activation=0
@error=0
@weights=[]
@prevUpdate=[]
for i in 0...numInputs
@weights.push(Useful.randomClamped)
@prevUpdate.push(0.0)
end
end
end
class NeuronLayer
attr_accessor:numNeurons
attr_accessor:neurons
=begin
public NeuronLayer(int _NumNeurons, int _NumInputsPerNeuron)
{
NumNeurons = _NumNeurons;
Neurons = new List<Neuron>();
for (int i = 0; i < NumNeurons; i++)
{
Neurons.Add(new Neuron(_NumInputsPerNeuron));
}
}
=end
def initialize(numNeurons,numInputsPerNeuron)
@numNeurons=numNeurons
@neurons=[]
#p @numNeurons
for i in 0...@numNeurons
@neurons.push(Neuron.new(numInputsPerNeuron))
end
end
end
class NeuralNet
attr_accessor:numInputs
attr_accessor:numOutputs
attr_accessor:numHiddenLayers
attr_accessor:neuronsPerHiddenLyr
attr_accessor:learningRate
attr_accessor:errorSum
attr_accessor:trained
attr_accessor:numEpochs
attr_accessor:layers
=begin
private bool NetworkTrainingEpoch(List<List<double>> SetIn, List<List<double>> SetOut)
{
if (Useful.WITH_MOMENTUM)
{
return NetworkTrainingEpochWithMomentum(SetIn, SetOut);
}
else
{
return NetworkTrainingEpochNonMomentum(SetIn, SetOut);
}
}
=end
def networkTrainingEpoch(setIn,setOut)
#if Useful::WITH_MOMENTUM
return networkTrainingEpochWithMomentum(setIn, setOut)
#else
# return networkTrainingEpochNonMomentum(SetIn, SetOut)
# end
end
def networkTrainingEpochWithMomentum(setIn,setOut)
@errorSum=0.0000000
for vec in 0...setIn.size
outputs=update(setIn[vec])
if outputs.size==0
p "Bad inputs"
return false
end
for op in 0...@numOutputs
err=0.000000000
err=(setOut[vec][op]-outputs[op])*outputs[op]*(1.0000000-outputs[op])
@layers[1].neurons[op].error=err
@errorSum+=(setOut[vec][op] - outputs[op]) * (setOut[vec][op] - outputs[op])
curWeight=0
curNrnHid=0
while (curWeight < @layers[1].neurons[op].weights.size - 1)
@layers[1].neurons[op].weights[curWeight] += err * @learningRate * @layers[0].neurons[curNrnHid].activation
curWeight+=1
curNrnHid+=1
end
@layers[1].neurons[op].weights[curWeight] += err * @learningRate * Useful::BIAS
end
curNrnHid=0
n=0
while (curNrnHid < @layers[0].neurons.size)
err=0.00000000
curNrnOut=0.00000000
while (curNrnOut < @layers[1].neurons.size)
err+=@layers[1].neurons[curNrnOut].error * @layers[1].neurons[curNrnOut].weights[n]
curNrnOut+=1
end
err*=@layers[0].neurons[curNrnHid].activation * (1.000000 - @layers[0].neurons[curNrnHid].activation)
for w in 0...@numInputs
@layers[0].neurons[curNrnHid].weights[w] += err * @learningRate * setIn[vec][w]
end
@layers[0].neurons[curNrnHid].weights[@numInputs] += err * @learningRate * Useful::BIAS
curNrnHid+=1
n+=1
end
end
return true
end
=begin
private void CreateNet()
{
if (NumHiddenLayers > 0)
{
//隐含层
Layers.Add(new NeuronLayer(NeuronsPerHiddenLyr, NumInputs));
for (int i = 0; i < NumHiddenLayers - 1; i++)
{
Layers.Add(new NeuronLayer(NeuronsPerHiddenLyr, NeuronsPerHiddenLyr));
}
//输出层
Layers.Add(new NeuronLayer(NumOutputs, NeuronsPerHiddenLyr));
}
else
{
//输出层
Layers.Add(new NeuronLayer(NumOutputs, NumInputs));
}
}
=end
def createNet
if @numHiddenLayers>0
#p @neuronsPerHiddenLyr
@layers.push(NeuronLayer.new(@neuronsPerHiddenLyr, @numInputs))
for i in 0...@numHiddenLayers -1
@layers.push(NeuronLayer.new(@neuronsPerHiddenLyr, @neuronsPerHiddenLyr))
end
@layers.push(NeuronLayer.new(@numOutputs, @neuronsPerHiddenLyr))
else
@layers.push(NeuronLayer.new(@numOutputs, @numInputs))
end
end
=begin
private void InitializeNetwork()
{
//对于每一层执行
for (int i = 0; i < NumHiddenLayers + 1; i++)
{
//对于每个神经元执行
for (int n = 0; n < Layers[i].NumNeurons; n++)
{
//对于每个权重执行
for (int k = 0; k < Layers[i].Neurons[n].NumInputs; k++)
{
Layers[i].Neurons[n].Weights[k] = Useful.RandomClamped();
}
}
}
ErrorSum = 9999;
NumEpochs = 0;
}
=end
def initializeNetwork
for i in 0...@numHiddenLayers+1
for n in 0...@layers[i].numNeurons #?
for k in 0...@layers[i].neurons[n].numInputs
@layers[i].neurons[n].weights[k] = Useful.randomClamped
end
end
end
@errorSum=9999
@numEpochs=0
end
=begin
private double Sigmoid(double activation, double response)
{
return (1.0 / (1.0 + Math.Exp(- activation / response)));
}
=end
def sigmoid(activation,response)
return (1.0000000/(1.0000000 + Math.exp(- activation / response)))
end
=begin
public NeuralNet(int _NumInputs, int _NumOutputs, int _HiddenNeurons, double _LearningRate)
{
NumInputs = _NumInputs;
NumOutputs = _NumOutputs;
NumHiddenLayers = 1;
NeuronsPerHiddenLyr = _HiddenNeurons;
LearningRate = _LearningRate;
ErrorSum = 9999;
Trained = false;
NumEpochs = 0;
Layers = new List<NeuronLayer>();
CreateNet();
}
=end
def initialize(numInputs,numOutputs,hiddenNeurons,learningRate)
@numInputs=numInputs
@numOutputs=numOutputs
@neuronsPerHiddenLyr=hiddenNeurons
@learningRate=learningRate
@numHiddenLayers=1
@errorSum=9999
@trained=false
@numEpochs=0
@layers=[]
createNet
end
=begin
public List<double> Update(List<double> _inputs)
{
List<double> inputs = new List<double>(_inputs);
List<double> outputs = new List<double>();
int cWeight = 0;
//验证输入长度
if (inputs.Count != NumInputs)
{
return outputs;
}
//对于每一层执行
for (int i = 0; i < NumHiddenLayers + 1; i++)
{
if (i > 0)
{
inputs = new List<double>(outputs);
}
outputs.Clear();
cWeight = 0;
=end
def update(inputs1)
inputs=inputs1.clone
outputs=[]
cWeight=0
if inputs.size!=@numInputs
return outputs
end
for i in 0...@numHiddenLayers+1
if (i>0)
inputs = outputs.clone
end
outputs.clear
cWeight=0
=begin
//对于每个神经元执行
for (int n = 0; n < Layers[i].NumNeurons; n++)
{
double netinput = 0;
int num = Layers[i].Neurons[n].NumInputs;
//对于每个权重执行
for (int k = 0; k < num - 1; k++)
{
netinput += Layers[i].Neurons[n].Weights[k] * inputs[cWeight++];
}
netinput += Layers[i].Neurons[n].Weights[num - 1] * Useful.BIAS;
Layers[i].Neurons[n].Activation = Sigmoid(netinput, Useful.ACTIVATION_RESPONSE);
outputs.Add(Layers[i].Neurons[n].Activation);
cWeight = 0;
}
}
return outputs;
}
=end
for n in 0...@layers[i].numNeurons
netinput=0.000000000
num=@layers[i].neurons[n].numInputs
for k in 0...num-1
netinput+=@layers[i].neurons[n].weights[k] * inputs[cWeight]
cWeight+=1
end
netinput += @layers[i].neurons[n].weights[num - 1] * Useful::BIAS
@layers[i].neurons[n].activation = sigmoid(netinput, Useful::ACTIVATION_RESPONSE)
outputs.push(layers[i].neurons[n].activation)
cWeight=0
end
end
return outputs
end
=begin
if ((SetIn.Count != SetOut.Count) || (SetIn[0].Count != NumInputs) || (SetOut[0].Count != NumOutputs))
{
throw new Exception("训练集输入输出不符！");
}
InitializeNetwork();
//训练直至错误小于阈值
while (ErrorSum > Useful.ERROR_THRESHOLD)
{
//迭代训练
if (!NetworkTrainingEpoch(SetIn, SetOut))
{
return false;
}
NumEpochs++;
//窗体刷新
SendMessage(NumEpochs, ErrorSum);
}
Trained = true;
return true;
=end
def train(data)
@v=Viewport.new(0,0,$原宽度,$原高度)
@v.z=99999999
@ss=Sprite.new
@ss.z=99999999
@ss.bitmap=Bitmap.new($原宽度,16)
@ss.bitmap.fill_rect(0,0,$原宽度,16,Color.new(255,255,255))
setIn=data.setIn.clone
setOut=data.setOut.clone
if ((setIn.size != setOut.size) or (setIn[0].size != @numInputs) or (setOut[0].size != @numOutputs))
print "训练集输入输出不符！"
exit(250)
end
initializeNetwork
t=0
while (@errorSum>Useful::ERROR_THRESHOLD)
if (!networkTrainingEpoch(setIn, setOut))
p "Bad Runnning"
return false
end
@numEpochs+=1
val=Useful::ERROR_THRESHOLD / @errorSum * $原宽度
@ss.bitmap.fill_rect(0,0,[[val,$原宽度].min,0].max.to_i,32,Color.new(0,0,255))
Graphics.update if (t % 100)==0
t+=1
end
@v.dispose
@ss.bitmap.dispose
@ss.dispose
@v=nil
@ss=nil
Graphics.update
Graphics.update
end
end
class GestureData
attr_accessor:names
attr_accessor:patterns
attr_accessor:patternNumber
attr_accessor:patternSize
attr_accessor:setIn
attr_accessor:setOut
def init
=begin
for (int j = 0; j < PatternNumber; j++)
{
List<double> temp = new List<double>();
for (int v = 0; v < PatternSize * 2; v++)
{
temp.Add(Useful.InitPatterns[j][v]);
}
Patterns.Add(temp);
Names.Add(Useful.InitNames[j]);
}
=end
for j in 0...@patternNumber
temp=[]
for v in 0...@patternSize*2
temp.push(Useful::InitPatterns[j][v])
end
@patterns.push(temp)
@names.push(Useful::InitNames[j])
end
end
=begin
public GestureData(int _PatternNumber, int _PatternSize)
{
Names = new List<string>();
Patterns = new List<List<double>>();
SetIn = new List<List<double>>();
SetOut = new List<List<double>>();
PatternNumber = _PatternNumber;
PatternSize = _PatternSize;
Init();
CreateTrainingSet();
}
=end
def initialize(patternNumber, patternSize)
@names=[]
@patterns=[]
@setIn=[]
@setOut=[]
@patternNumber=patternNumber
@patternSize=patternSize
init
createTrainingSet
end
=begin
public string PatternName(int index)
{
if (Names[index] != null)
{
return Names[index];
}
else
{
return "";
}
}
=end
def patternName(index)
if @names[index]!=nil
return @names[index]
else
return ""
end
end
=begin
public bool AddPattern(List<double> _Pattern, string _Name)
{
//检查手势向量长度
if (_Pattern.Count != PatternSize * 2)
{
throw new Exception("手势向量长度错误！");
}
Names.Add(_Name);
Patterns.Add(new List<double>(_Pattern));
PatternNumber++;
CreateTrainingSet();
return true;
}
=end
def addPattern(pattern, name)
if pattern.size!=@patternSize*2
print "手势向量长度错误！"
exit
end
@names.push(name)
@pattern.push(pattern.clone)
@patternNumber+=1
createTrainingSet
return true
end
=begin
public void CreateTrainingSet()
{
//清空训练集
SetIn.Clear();
SetOut.Clear();
//对每个手势操作
for (int j = 0; j < PatternNumber; j++)
{
SetIn.Add(Patterns[j]);
//相关的输出为1，不相关的输出为0
List<double> outputs = new List<double>();
for (int i = 0; i < PatternNumber; i++)
{
outputs.Add(0);
}
outputs[j] = 1;
SetOut.Add(outputs);
}
}
=end
def createTrainingSet
setIn.clear
setOut.clear
for j in 0...@patternNumber
setIn.push(@patterns[j])
outputs=[]
for i in 0...@patternNumber
outputs.push(0)
end
outputs[j]=1
setOut.push(outputs)
end
end
end
=begin
private void RenewNetwork()
{
Useful.LEARNING_RATE = (double)(txtLearning.Value);
Useful.ERROR_THRESHOLD = (double)(txtThreshold.Value);
Useful.NUM_HIDDEN_NEURONS = (int)(txtHidden.Value);
Useful.WITH_MOMENTUM = chkMomentum.Checked;
Useful.MOMENTUM = (double)(txtMomentum.Value);
Useful.WITH_NOISE = chkNoise.Checked;
Useful.MAX_NOISE_TO_ADD = (double)(txtNoise.Value);
net = new NeuralNet(Useful.NUM_VECTORS * 2, NumValidPatterns, Useful.NUM_HIDDEN_NEURONS, Useful.LEARNING_RATE);
net.SendMessage += new NeuralNet.DelegateOfSendMessage(ShowMessage);
txtState.Text = "Training";
TrainNetwork();
txtState.Text = "Ready";
}
private bool TrainNetwork()
{
Mode = RUN_MODE.TRAINING;
if (!(net.Train(data)))
{
return false;
}
Mode = RUN_MODE.ACTIVE;
return true;
}
=end
module NetworkFunctions
=begin
//手势相关数据对象
private GestureData data;
//神经网络对象
private NeuralNet net;
//手势数目
private int NumValidPatterns;
//需要记录的鼠标点数
private int NumSmoothPoints;
//用户鼠标输入的手势向量
private List<Point> RawPath;
//光滑化之后的手势向量
private List<Point> SmoothPath;
//待匹配的向量
private List<double> Vectors;
//网络最大的输出（最像的匹配）
private double HighestOutput;
//网络最大的输出对应的手势
private int BestMatch;
//匹配的手势
private int Match;
//程序的运行状态
private RUN_MODE Mode;
=end
attr_accessor:net
attr_accessor:data
attr_accessor:mode
attr_accessor:numValidPatterns
attr_accessor:numSmoothPoints
attr_accessor:rawPath
attr_accessor:smoothPath
attr_accessor:vectors
attr_accessor:highestOutput
attr_accessor:bestMatch
attr_accessor:match
def self.net
return @net
end
def self.data
return @data
end
def self.numValidPatterns
return @numValidPatterns
end
def self.numSmoothPoints
return @numSmoothPoints
end
def self.rawPath
return @rawPath
end
def self.smoothPath
return @smoothPath
end
def self.vectors
return @vectors
end
def self.highestOutput
return @highestOutput
end
def self.bestMatch
return @bestMatch
end
def self.match
return @match
end
def self.net=(v)
return @net=v
end
def self.data=(v)
return @data=v
end
def self.numValidPatterns=(v)
return @numValidPatterns=v
end
def self.numSmoothPoints=(v)
return @numSmoothPoints=v
end
def self.rawPath=(v)
return @rawPath=v
end
def self.smoothPath=(v)
return @smoothPath=v
end
def self.vectors=(v)
return @vectors=v
end
def self.highestOutput=(v)
return @highestOutput=v
end
def self.bestMatch=(v)
return @bestMatch=v
end
def self.match=(v)
return @match=v
end
def self.renewNetwork
@net=NeuralNet.new(Useful::NUM_VECTORS * 2, @numValidPatterns, Useful::NUM_HIDDEN_NEURONS, Useful::LEARNING_RATE)
trainNetwork
return true
end
def self.trainNetwork
@mode=Useful::RUN_MODE::TRAINING
if !(@net.train(@data))
return false
end
@mode=Useful::RUN_MODE::ACTIVE
end
def self.mode
return @mode
end
def self.mode=(v)
return @mode=v
end
def self.initData
@mode=Useful::RUN_MODE::UNREADY
@numValidPatterns=Useful::NUM_PATTERNS
@numSmoothPoints=Useful::NUM_VECTORS + 1
@highestOutput=0.00000000
@bastMatch=-1
@match=-1
@rawPath=[]
@smoothPath=[]
@vectors=[]
@data=GestureData.new(@numValidPatterns, Useful::NUM_VECTORS)
@net=NeuralNet.new(Useful::NUM_VECTORS * 2, @numValidPatterns, Useful::NUM_HIDDEN_NEURONS, Useful::LEARNING_RATE)
trainNetwork
end
def self.save
save_data(@net,"BPNet.rvdata")
save_data(@data,"BPNetD.rvdata")
end
def self.load
@net = load_data("BPNet.rvdata")
@data= load_data("BPNetD.rvdata")
@mode=Useful::RUN_MODE::ACTIVE
@numValidPatterns=Useful::NUM_PATTERNS
@numSmoothPoints=Useful::NUM_VECTORS + 1
@highestOutput=0.00000000
@bastMatch=-1
@match=-1
@rawPath=[]
@smoothPath=[]
@vectors=[]
end
end
=begin
private void InitData()
{
Mode = RUN_MODE.UNREADY;
NumValidPatterns = Useful.NUM_PATTERNS;
NumSmoothPoints = Useful.NUM_VECTORS + 1;
HighestOutput = 0.0;
BestMatch = -1;
Match = -1;
RawPath = new List<Point>();
SmoothPath = new List<Point>();
Vectors = new List<double>();
data = new GestureData(NumValidPatterns, Useful.NUM_VECTORS);
net = new NeuralNet(Useful.NUM_VECTORS * 2, NumValidPatterns, Useful.NUM_HIDDEN_NEURONS, Useful.LEARNING_RATE);
net.SendMessage += new NeuralNet.DelegateOfSendMessage(ShowMessage);
}
=end
#NetworkFunctions.initData
NetworkFunctions.load
#NetworkFunctions.load

复制代码

class Bitmap
#--------------------------------------------------------------------------
# ● 描绘直线
# x1，y1，x2，y2：直线两端的坐标
# width：宽度
# color：颜色
#--------------------------------------------------------------------------
def drawline(x1, y1, x2, y2, width, color)
x1 = x1.to_f
y1 = y1.to_f
x2 = x2.to_f
y2 = y2.to_f
width = width.to_f
k = (y2 - y1) / (x2 - x1)
if k.abs > 1
drawline_x(x1, y1, x2, y2, width, color)
else
drawline_y(x1, y1, x2, y2, width, color)
end
end
def drawline_x(x1, y1, x2, y2, width, color)
l = ((x1 - x2) ** 2 + (y1 - y2) ** 2) ** 0.5 * width / (y1 - y2)
length = l.abs * 2
k = (x2 - x1) / (y2 - y1) #x=ky+b
b = x1 - k * y1
if l > 0
for ty in y2.to_i..y1.to_i
tx = ty * k + b
fill_rect(tx - l, ty, length, 1, color)
end
else
for ty in y1.to_i..y2.to_i
tx = ty * k + b
fill_rect(tx + l, ty, length, 1, color)
end
end
end
def drawline_y(x1, y1, x2, y2, width, color)
l = ((x1 - x2) ** 2 + (y1 - y2) ** 2) ** 0.5 * width / (x1 - x2)
height = l.abs * 2
k = (y2 - y1) / (x2 - x1) #y=kx+b
b = y1 - k * x1
if l > 0
for tx in x2.to_i..x1.to_i
ty = tx * k + b
fill_rect(tx, ty - l, 1, height, color)
end
else
for tx in x1.to_i..x2.to_i
ty = tx * k + b
fill_rect(tx, ty + l, 1, height, color)
end
end
end
end
class DrawBox
include NetworkFunctions
def initialize(x,y,width,height)
@draw_sprite=Sprite.new
@draw_sprite.x=x
@draw_sprite.y=y
@draw_sprite.z=999999999
@draw_sprite.bitmap=Bitmap.new(width,height)
NetworkFunctions.rawPath=[]
@draw_area=Rect.new(x,y,width,height)
@in_drawing=false
end
def update
return if not Mouse.rect?(@draw_area)
if Mouse.up?(1)
# mouse_up
@in_drawing=false
@draw_sprite.bitmap.clear
startMatch
end
if Mouse.click?(1)
# mouse_down
@draw_sprite.bitmap.clear
@in_drawing=true
@ox,@oy=Mouse.pos
NetworkFunctions.rawPath.clear
NetworkFunctions.smoothPath.clear
NetworkFunctions.vectors.clear
NetworkFunctions.rawPath.push(Mouse.pos)
end
if Mouse.pos!=[@ox,@oy] and @in_drawing
@draw_sprite.bitmap.drawline(@ox,@oy, Mouse.pos[0],Mouse.pos[1],3,Color.new(255,255,255,88))
NetworkFunctions.rawPath.push(Mouse.pos)
@ox,@oy=Mouse.pos
end
end
def startMatch
if smooth
createVectors
if (NetworkFunctions.mode == Useful::RUN_MODE::ACTIVE)
if not testForMatch
return false
else
return showResult
end
else
#p "bye"
end
end
end
def showResult
p NetworkFunctions.data.patternName(NetworkFunctions.bestMatch)
end
=begin
private bool TestForMatch()
{
List<double> outputs = net.Update(Vectors);
if (outputs.Count == 0)
{
return false;
}
HighestOutput = 0;
BestMatch = 0;
Match = -1;
for (int i = 0; i < outputs.Count; i++)
{
if (outputs[i] > HighestOutput)
{
//记录最像的匹配
HighestOutput = outputs[i];
BestMatch = i;
//确定是这个手势
if (HighestOutput > Useful.MATCH_TOLERANCE)
{
Match = BestMatch;
}
}
}
return true;
}
=end
def testForMatch
outputs=NetworkFunctions.net.update(NetworkFunctions.vectors)
if outputs.size==0
return false
end
NetworkFunctions.highestOutput=0
NetworkFunctions.bestMatch=0
NetworkFunctions.match=-1
for i in 0...outputs.size
if outputs[i]>NetworkFunctions.highestOutput
NetworkFunctions.highestOutput=outputs[i]
NetworkFunctions.bestMatch=i
if (NetworkFunctions.highestOutput>Useful::MATCH_TOLERANCE)
NetworkFunctions.match=NetworkFunctions.bestMatch
end
end
end
return true
end
=begin
private bool Smooth()
{
//确保包含计算所需的足够的点
if (RawPath.Count < NumSmoothPoints)
{
return false;
}
SmoothPath = new List<Point>(RawPath);
//对所有的最小跨度点对取中点，删除原来的点，循环执行
while (SmoothPath.Count > NumSmoothPoints)
{
double ShortestSoFar = double.MaxValue;
int PointMarker = 0;
//计算最小跨度
for (int SpanFront = 2; SpanFront < SmoothPath.Count - 1; SpanFront++)
{
//计算点对距离
double length = Math.Sqrt((double)
((SmoothPath[SpanFront - 1].X - SmoothPath[SpanFront].X) *
(SmoothPath[SpanFront - 1].X - SmoothPath[SpanFront].X) +
(SmoothPath[SpanFront - 1].Y - SmoothPath[SpanFront].Y) *
(SmoothPath[SpanFront - 1].Y - SmoothPath[SpanFront].Y)));
if (length < ShortestSoFar)
{
ShortestSoFar = length;
PointMarker = SpanFront;
}
}
//插入中点，删除原来的点
Point newPoint = new Point();
newPoint.X = (SmoothPath[PointMarker - 1].X + SmoothPath[PointMarker].X) / 2;
newPoint.Y = (SmoothPath[PointMarker - 1].Y + SmoothPath[PointMarker].Y) / 2;
SmoothPath[PointMarker - 1] = newPoint;
SmoothPath.RemoveAt(PointMarker);
}
return true;
}
=end
def smooth
if NetworkFunctions.rawPath.size<NetworkFunctions.numSmoothPoints
return false
end
NetworkFunctions.smoothPath=NetworkFunctions.rawPath.clone
while (NetworkFunctions.smoothPath.size>NetworkFunctions.numSmoothPoints)
shortestSoFar=1.79769313486232E+308
pointMarker=0
for spanFront in 2...NetworkFunctions.smoothPath.size-1
length=Math.sqrt(((NetworkFunctions.smoothPath[spanFront - 1][0] - NetworkFunctions.smoothPath[spanFront][0]) *
(NetworkFunctions.smoothPath[spanFront - 1][0] - NetworkFunctions.smoothPath[spanFront][0]) +
(NetworkFunctions.smoothPath[spanFront - 1][1] - NetworkFunctions.smoothPath[spanFront][1]) *
(NetworkFunctions.smoothPath[spanFront - 1][1] - NetworkFunctions.smoothPath[spanFront][1])))
if length<shortestSoFar
shortestSoFar=length
pointMarker=spanFront
end
end
newPoint=[]
newPoint[0]=(NetworkFunctions.smoothPath[pointMarker - 1][0] + NetworkFunctions.smoothPath[pointMarker][0]) / 2.0000000000
newPoint[1]=(NetworkFunctions.smoothPath[pointMarker - 1][1] + NetworkFunctions.smoothPath[pointMarker][1]) / 2.0000000000
NetworkFunctions.smoothPath[pointMarker-1]=newPoint
NetworkFunctions.smoothPath.delete_at(pointMarker)
end
return true
end
=begin
for (int p = 1; p < SmoothPath.Count; ++p)
{
double x = (double)(SmoothPath[p].X - SmoothPath[p - 1].X);
double y = (double)(SmoothPath[p].Y - SmoothPath[p - 1].Y);
double len = Math.Sqrt((double)(x * x + y * y));
Vectors.Add(x / len);
Vectors.Add(y / len);
}
=end
def createVectors
for p in 1...NetworkFunctions.smoothPath.size
x=(NetworkFunctions.smoothPath[p][0] - NetworkFunctions.smoothPath[p - 1][0]).to_f
y=(NetworkFunctions.smoothPath[p][1] - NetworkFunctions.smoothPath[p - 1][1]).to_f
len=Math.sqrt(x*x+y*y).to_f
NetworkFunctions.vectors.push(x/len)
NetworkFunctions.vectors.push(y/len)
end
end
end

复制代码

手势文件放到根目录（可调）

Net.rar (4.52 KB, 下载次数: 248)

光的圆周率 · 发表于 2010-12-31 22:02:11

强大啊

1049079876 · 发表于 2011-1-3 06:47:08

提示: 作者被禁止或删除内容自动屏蔽

沉影不器 · 发表于 2011-1-4 21:51:05

提示: 作者被禁止或删除内容自动屏蔽

yangff · 发表于 2011-1-11 21:21:50

沉影不器发表于 2011-1-4 21:51
这篇C++正好看过...
当时的感想是:如果只做到像很多IE浏览器那样鼠标手势的话,倒是容易得多,就不需要神经了 ...

识别率和效果问题。
只是要上下左右就不用神经了必须的……

帐号		自动登录	找回密码
密码			注册会员

[RMVX发布] 传说中的BP手势识别

加入我们，或者，欢迎回来。

点评

烫烫烫

开拓者

第1届Title华丽大赛新人奖