javaCV视频处理技术＞提取人像视频|百度AI

效果图对比

 左侧的为原视频，右侧为提取人像跳舞的视频。

之前写的文章 JAVA代码实现人物照片的人像分割 | 百度AI 是处理图片的 ，视频处理也是在图片处理基础上实现的。

pom文件引入依赖



com.baidu.aip
java-sdk
4.16.3



org.bytedeco
javacv-platform
1.5.5


 java核心实现代码（完整）


import com.baidu.aip.bodyanalysis.AipBodyAnalysis;
import org.bytedeco.javacv.FFmpegFrameGrabber;
import org.bytedeco.javacv.FFmpegFrameRecorder;
import org.bytedeco.javacv.Frame;
import org.bytedeco.javacv.Java2DFrameConverter;
import javax.imageio.ImageIO;
import java.awt.*;
import java.awt.image.BufferedImage;
import java.io.*;
import java.util.HashMap;
import org.bytedeco.ffmpeg.global.avutil;
import org.bytedeco.ffmpeg.global.avcodec;
import org.json.JSONObject;
import sun.misc.BASE64Decoder;
public class VideoProcessor {
//设置APPID/AK/SK
public static final String APP_ID = "25393592";
public static final String API_KEY = "OkRDD6FQwm5hTKGSMIEL9RN4";
public static final String SECRET_KEY = "ONAxohflnqL2HwBEQB2iGUCjmO5lgywp";
final static String videoFolderPath = "C:/Users/liuya/Desktop/video/";
final static String videOname= "demo.mp4";
final static String imageFolderPath = "C:/Users/liuya/Desktop/people/";
public static void main(String[] args) throws Exception {
videoProcess(videoFolderPath + videoName);
}
//视频水印
public static void videoProcess(String filePath) {
//抓取视频图像资源
FFmpegFrameGrabber videoGrabber = new FFmpegFrameGrabber(filePath);
//抓取视频图像资源
FFmpegFrameGrabber audioGrabber = new FFmpegFrameGrabber(filePath);
try {
videoGrabber.start();
audioGrabber.start();
FFmpegFrameRecorder recorder = new FFmpegFrameRecorder(videoFolderPath + "new" + videoName, videoGrabber.getImageWidth(), videoGrabber.getImageHeight(), videoGrabber.getAudioChannels());
recorder.setPixelFormat(avutil.AV_PIX_FMT_YUV420P);
recorder.setVideoCodec(avcodec.AV_CODEC_ID_H264);
recorder.start();
//处理图像
int videoSize = videoGrabber.getLengthInVideoFrames();
for (int i = 0; i Frame videoFrame = videoGrabber.grabImage();
if (videoFrame != null && videoFrame.image != null) {
System.out.println("视频共" + videoSize + "帧，正处理第" + (i + 1) + "帧图片");
Java2DFrameConverter cOnverter= new Java2DFrameConverter();
BufferedImage bi=converter.getBufferedImage(videoFrame);
BufferedImage bufferedImage = splitting(bi);
recorder.record(converter.convert(bufferedImage));
}
}
//处理音频
for (int i = 0; i Frame audioFrame = audioGrabber.grabSamples();
if (audioFrame != null && audioFrame.samples != null) {
recorder.recordSamples(audioFrame.sampleRate, audioFrame.audioChannels, audioFrame.samples);
}
}
recorder.stop();
recorder.release();
videoGrabber.stop();
audioGrabber.stop();
} catch (Exception e) {
e.printStackTrace();
}
}
public static BufferedImage splitting(BufferedImage image){
ByteArrayOutputStream out=new ByteArrayOutputStream();
try {
ImageIO.write(image,"png",out);
} catch (IOException e) {
e.printStackTrace();
}
return splitting(out.toByteArray());
}
public static BufferedImage splitting(byte[] image){
// 初始化一个AipBodyAnalysis
AipBodyAnalysis client = new AipBodyAnalysis(APP_ID, API_KEY, SECRET_KEY);
// 可选：设置网络连接参数
client.setConnectionTimeoutInMillis(2000);
client.setSocketTimeoutInMillis(60000);
// 传入可选参数调用接口
HashMap optiOns= new HashMap();
options.put("type", "foreground");
// 参数为本地路径
JSONObject res = client.bodySeg(image, options);
return convert(res.get("foreground").toString());
}
public static BufferedImage convert(String labelmapBase64) {
try {
BASE64Decoder decoder = new BASE64Decoder();
byte[] bytes = decoder.decodeBuffer(labelmapBase64);
InputStream is = new ByteArrayInputStream(bytes);
BufferedImage image = ImageIO.read(is);
//失真处理
BufferedImage newBufferedImage = new BufferedImage(image.getWidth(), image.getHeight(), BufferedImage.TYPE_INT_RGB);
newBufferedImage.createGraphics().drawImage(image, 0, 0, Color.WHITE, null);
ByteArrayOutputStream out=new ByteArrayOutputStream();
ImageIO.write(newBufferedImage, "png", out);
ByteArrayInputStream in = new ByteArrayInputStream(out.toByteArray());
return ImageIO.read(in);
} catch (IOException e) {
e.printStackTrace();
return null;
}
}
}


控制台输出

文章到此为止，如果疑问，评论区留言！！！

觉得不错，就点赞收藏吧！！！