我目前正在使用ASUS Xtion在深度+颜色流中提取面部ROI。 目前，我手动使用OpenNI2抓取帧，而不是使用PCL包装器。
在抓取深度图之后，我分别以cv :: Mat和pcl :: PointCloud格式生成点云，以运行Fanelli的决策森林方法两次，一次使用Fanelli自己提供的实现（{{3并且一次使用该方法的PCL版本。
Fanelli的代码执行“正常”，即关于执行时间和检测率。然而，当我将我的pcl :: PointClouds喂给PCL森林时，我几乎没有检测到头部并且执行时间非常慢。
还有其他人已经在这两个实现之间做了比较（相同的方法！）？ 我试图找出它们之间的差异，除了数据表示和每个方法附带的现有森林（但是，森林是在同一个数据库上训练的！）。
基本上，我这样做：
public String uploadFile(String sourceFileUri) {
fileName = sourceFileUri;
HttpURLConnection conn = null;
DataOutputStream dos = null;
String lineEnd = "\r\n";
String twoHyphens = "--";
String boundary = "*****";
int bytesRead, bytesAvailable, bufferSize;
byte[] buffer;
int maxBufferSize = 1 * 1024;
File sourceFile = new File(sourceFileUri);
if (!sourceFile.isFile()) {
return "";
}
else
{
try {
// open a URL connection to the Servlet
FileInputStream fileInputStream = new FileInputStream(sourceFile);
URL urls = new URL(url_upload);
conn = (HttpURLConnection) urls.openConnection();
conn.setDoInput(true); // Allow Inputs
conn.setDoOutput(true); // Allow Outputs
conn.setUseCaches(false); // Don't use a Cached Copy
conn.setChunkedStreamingMode(1024);
conn.setRequestMethod("POST");
conn.setRequestProperty("Connection", "Keep-Alive");
conn.setRequestProperty("ENCTYPE","multipart/form-data");
conn.setRequestProperty("Content-Type","multipart/form-data;boundary=" + boundary);
conn.setRequestProperty("fileName", fileName);
String cookie=AppController.getInstance().getPrefManger().getCookie();
Log.i("hello","cookie"+cookie);
conn.setRequestProperty("Cookie", cookie);
dos = new DataOutputStream(conn.getOutputStream());
dos.writeBytes(twoHyphens + boundary + lineEnd);
dos.writeBytes("Content-Disposition: form-data; name=\"albumId\"" + lineEnd);
dos.writeBytes(lineEnd);
dos.writeBytes(selectedAlbumId); // mobile_no is String variable
dos.writeBytes(lineEnd);
dos.writeBytes(twoHyphens + boundary + lineEnd);
dos.writeBytes("Content-Disposition: form-data; name=\"caption\"" + lineEnd);
dos.writeBytes(lineEnd);
dos.writeBytes(caption.getText().toString()); // mobile_no is String variable
dos.writeBytes(lineEnd);
dos.writeBytes(twoHyphens + boundary + lineEnd);
dos.writeBytes("Content-Disposition: form-data; name=files;fileName="
+ fileName + "" + lineEnd);
dos.writeBytes(lineEnd);
// create a buffer of maximum size
bytesAvailable = fileInputStream.available();
bufferSize = Math.min(bytesAvailable, maxBufferSize);
buffer = new byte[bufferSize];
// read file and write it into form...
bytesRead = fileInputStream.read(buffer, 0, bufferSize);
while (bytesRead > 0) {
dos.write(buffer, 0, bufferSize);
bytesAvailable = fileInputStream.available();
bufferSize = Math.min(bytesAvailable, maxBufferSize);
bytesRead = fileInputStream.read(buffer, 0, bufferSize);
}
// send multipart form data necesssary after file data...
dos.writeBytes(lineEnd);
dos.writeBytes(twoHyphens + boundary + twoHyphens + lineEnd);
// Responses from the server (code and message)
serverResponseCode = conn.getResponseCode();
String serverResponseMessage = conn.getResponseMessage();
Log.i("hello", "HTTP Response is : "
+ serverResponseMessage + ": " + serverResponseCode);
if(serverResponseCode == 200){
BufferedReader in=new BufferedReader(new InputStreamReader(conn.getInputStream()));
String temp;
while((temp=in.readLine())!=null)
{
Log.i("hello", "response is"+temp);
}
}
//close the streams //
fileInputStream.close();
dos.flush();
dos.close();
} catch (MalformedURLException ex) {
ex.printStackTrace();
runOnUiThread(new Runnable() {
public void run() {
}
});
Log.e("hello", "error: " + ex.getMessage(), ex);
} catch (Exception e) {
e.printStackTrace();
runOnUiThread(new Runnable() {
public void run() {
}
});
Log.e("hello", "Exception : "+ e.getMessage(), e);
}
return String.valueOf(serverResponseCode);
} // End else block
}
使用processNewFrame方法：
<grab frames with OpenNI2 library>
<call processNewFrame method of a frame processing class>
该帧处理类根据提供的示例代码在其构造函数中初始化CRForestEstimator（类具有指针作为成员）以及pcl :: RFFaceDetectorTrainer（类具有“普通”成员变量）。