资讯 社区 文档
技术能力
语音技术
文字识别
人脸与人体
图像技术
语言与知识
视频技术

人体关键点识别

接口描述

对于输入的一张图片(可正常解码,且长宽比适宜),检测图片中的所有人体,输出每个人体的21个主要关键点,包含头顶、五官、脖颈、四肢等部位,同时输出人体的坐标信息和数量

支持多人检测、人体位置重叠、遮挡、背面、侧面、中低空俯拍、大动作等复杂场景。

21个关键点的位置:头顶、左耳、右耳、左眼、右眼、鼻子、左嘴角、右嘴角、脖子、左肩、右肩、左手肘、右手肘、左手腕、右手腕、左髋部、右髋部、左膝、右膝、左脚踝、右脚踝。示意图如下,正在持续扩展更多关键点,敬请期待。

单人场景:

多人场景:

在线调试

您可以在 示例代码中心 中调试该接口,可进行签名验证、查看在线调用的请求内容和返回结果、示例代码的自动生成。

请求说明

请求示例

HTTP 方法:POST

请求URL: https://aip.baidubce.com/rest/2.0/image-classify/v1/body_analysis

URL参数:

参数
access_token 通过API Key和Secret Key获取的access_token,参考“Access Token获取

Header如下:

参数
Content-Type application/x-www-form-urlencoded

Body中放置请求参数,参数详情如下:

请求参数

参数 是否必选 类型 可选值范围 说明
image string - 图像数据,base64编码后进行urlencode,要求base64编码和urlencode后大小不超过4M。图片的base64编码是不包含图片头的,如(data:image/jpg;base64,),支持图片格式:jpg、bmp、png,最短边至少50px,最长边最大4096px

请求代码示例

提示一:使用示例代码前,请记得替换其中的示例Token、图片地址或Base64信息。

提示二:部分语言依赖的类或库,请在代码注释中查看下载地址。

人体关键点识别
curl -i -k 'https://aip.baidubce.com/rest/2.0/image-classify/v1/body_analysis?access_token=【调用鉴权接口获取的token】' --data 'image=【图片Base64编码,需UrlEncode】' -H 'Content-Type:application/x-www-form-urlencoded'
<?php
/**
 * 发起http post请求(REST API), 并获取REST请求的结果
 * @param string $url
 * @param string $param
 * @return - http response body if succeeds, else false.
 */
function request_post($url = '', $param = '')
{
    if (empty($url) || empty($param)) {
        return false;
    }

    $postUrl = $url;
    $curlPost = $param;
    // 初始化curl
    $curl = curl_init();
    curl_setopt($curl, CURLOPT_URL, $postUrl);
    curl_setopt($curl, CURLOPT_HEADER, 0);
    // 要求结果为字符串且输出到屏幕上
    curl_setopt($curl, CURLOPT_RETURNTRANSFER, 1);
    curl_setopt($curl, CURLOPT_SSL_VERIFYPEER, false);
    // post提交方式
    curl_setopt($curl, CURLOPT_POST, 1);
    curl_setopt($curl, CURLOPT_POSTFIELDS, $curlPost);
    // 运行curl
    $data = curl_exec($curl);
    curl_close($curl);

    return $data;
}

$token = '[调用鉴权接口获取的token]';
$url = 'https://aip.baidubce.com/rest/2.0/image-classify/v1/body_analysis?access_token=' . $token;
$img = file_get_contents('[本地文件路径]');
$img = base64_encode($img);
$bodys = array(
    'image' => $img
);
$res = request_post($url, $bodys);

var_dump($res);~~~~
package com.baidu.ai.aip;

import com.baidu.ai.aip.utils.Base64Util;
import com.baidu.ai.aip.utils.FileUtil;
import com.baidu.ai.aip.utils.HttpUtil;

import java.net.URLEncoder;

/**
* 人体关键点识别
*/
public class BodyAnalysis {

    /**
    * 重要提示代码中所需工具类
    * FileUtil,Base64Util,HttpUtil,GsonUtils请从
    * https://ai.baidu.com/file/658A35ABAB2D404FBF903F64D47C1F72
    * https://ai.baidu.com/file/C8D81F3301E24D2892968F09AE1AD6E2
    * https://ai.baidu.com/file/544D677F5D4E4F17B4122FBD60DB82B3
    * https://ai.baidu.com/file/470B3ACCA3FE43788B5A963BF0B625F3
    * 下载
    */
    public static String body_analysis() {
        // 请求url
        String url = "https://aip.baidubce.com/rest/2.0/image-classify/v1/body_analysis";
        try {
            // 本地文件路径
            String filePath = "[本地文件路径]";
            byte[] imgData = FileUtil.readFileByBytes(filePath);
            String imgStr = Base64Util.encode(imgData);
            String imgParam = URLEncoder.encode(imgStr, "UTF-8");

            String param = "image=" + imgParam;

            // 注意这里仅为了简化编码每一次请求都去获取access_token,线上环境access_token有过期时间, 客户端可自行缓存,过期后重新获取。
            String accessToken = "[调用鉴权接口获取的token]";

            String result = HttpUtil.post(url, accessToken, param);
            System.out.println(result);
            return result;
        } catch (Exception e) {
            e.printStackTrace();
        }
        return null;
    }

    public static void main(String[] args) {
        BodyAnalysis.body_analysis();
    }
}
# encoding:utf-8

import requests
import base64

'''
人体关键点识别
'''

request_url = "https://aip.baidubce.com/rest/2.0/image-classify/v1/body_analysis"
# 二进制方式打开图片文件
f = open('[本地文件]', 'rb')
img = base64.b64encode(f.read())

params = {"image":img}
access_token = '[调用鉴权接口获取的token]'
request_url = request_url + "?access_token=" + access_token
headers = {'content-type': 'application/x-www-form-urlencoded'}
response = requests.post(request_url, data=params, headers=headers)
if response:
    print (response.json())
#include <iostream>
#include <curl/curl.h>

// libcurl库下载链接:https://curl.haxx.se/download.html
// jsoncpp库下载链接:https://github.com/open-source-parsers/jsoncpp/
const static std::string request_url = "https://aip.baidubce.com/rest/2.0/image-classify/v1/body_analysis";
static std::string body_analysis_result;
/**
 * curl发送http请求调用的回调函数,回调函数中对返回的json格式的body进行了解析,解析结果储存在全局的静态变量当中
 * @param 参数定义见libcurl文档
 * @return 返回值定义见libcurl文档
 */
static size_t callback(void *ptr, size_t size, size_t nmemb, void *stream) {
    // 获取到的body存放在ptr中,先将其转换为string格式
    body_analysis_result = std::string((char *) ptr, size * nmemb);
    return size * nmemb;
}
/**
 * 人体关键点识别
 * @return 调用成功返回0,发生错误返回其他错误码
 */
int body_analysis(std::string &json_result, const std::string &access_token) {
    std::string url = request_url + "?access_token=" + access_token;
    CURL *curl = NULL;
    CURLcode result_code;
    int is_success;
    curl = curl_easy_init();
    if (curl) {
        curl_easy_setopt(curl, CURLOPT_URL, url.data());
        curl_easy_setopt(curl, CURLOPT_POST, 1);
        curl_httppost *post = NULL;
        curl_httppost *last = NULL;
        curl_formadd(&post, &last, CURLFORM_COPYNAME, "image", CURLFORM_COPYCONTENTS, "【base64_img】", CURLFORM_END);

        curl_easy_setopt(curl, CURLOPT_HTTPPOST, post);
        curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, callback);
        result_code = curl_easy_perform(curl);
        if (result_code != CURLE_OK) {
            fprintf(stderr, "curl_easy_perform() failed: %s\n",
                    curl_easy_strerror(result_code));
            is_success = 1;
            return is_success;
        }
        json_result = body_analysis_result;
        curl_easy_cleanup(curl);
        is_success = 0;
    } else {
        fprintf(stderr, "curl_easy_init() failed.");
        is_success = 1;
    }
    return is_success;
}
using System;
using System.IO;
using System.Net;
using System.Text;
using System.Web;

namespace com.baidu.ai
{
    public class BodyAnalysis
    {
        // 人体关键点识别
        public static string body_analysis()
        {
            string token = "[调用鉴权接口获取的token]";
            string host = "https://aip.baidubce.com/rest/2.0/image-classify/v1/body_analysis?access_token=" + token;
            Encoding encoding = Encoding.Default;
            HttpWebRequest request = (HttpWebRequest)WebRequest.Create(host);
            request.Method = "post";
            request.KeepAlive = true;
            // 图片的base64编码
            string base64 = getFileBase64("[本地图片文件]");
            String str = "image=" + HttpUtility.UrlEncode(base64);
            byte[] buffer = encoding.GetBytes(str);
            request.ContentLength = buffer.Length;
            request.GetRequestStream().Write(buffer, 0, buffer.Length);
            HttpWebResponse response = (HttpWebResponse)request.GetResponse();
            StreamReader reader = new StreamReader(response.GetResponseStream(), Encoding.Default);
            string result = reader.ReadToEnd();
            Console.WriteLine("人体关键点识别:");
            Console.WriteLine(result);
            return result;
        }

        public static String getFileBase64(String fileName) {
            FileStream filestream = new FileStream(fileName, FileMode.Open);
            byte[] arr = new byte[filestream.Length];
            filestream.Read(arr, 0, (int)filestream.Length);
            string baser64 = Convert.ToBase64String(arr);
            filestream.Close();
            return baser64;
        }
    }
}

返回说明

接口除了返回人体框和每个关键点的坐标信息外,还会输出人体框和关键点的概率分数,实际应用中可以基于概率分数进行过滤,排除掉分数低的误识别“无效人体”推荐的过滤方案:当关键点得分大于0.2的个数大于3,且人体框的得分大于0.03时,才认为是有效人体

实际应用中,可根据对误识别、漏识别的容忍程度,调整阈值过滤方案,灵活应用,比如对误识别容忍低的应用场景,人体框的得分阈值可以提到0.06甚至更高。

返回参数

字段 是否必选 类型 说明
log_id uint64 唯一的log id,用于问题定位
person_num uint32 人体数目
person_info object[] 人体姿态信息
+location object 人体坐标信息
++height float 人体区域的高度
++left float 人体区域离左边界的距离
++top float 人体区域离上边界的距离
++width float 人体区域的宽度
++score float 人体框的概率分数,取值0-1,得分越接近1表示识别准确的概率越大
+body_parts object 身体部位信息,包含21个关键点
++top_head object 头顶
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++left_eye object 左眼
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++right_eye object 右眼
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++nose object 鼻子
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++left_ear object 左耳
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++right_ear object 右耳
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++left_mouth_corner object 左嘴角
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++right_mouth_corner object 右嘴角
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++neck object 颈部
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++left_shoulder object 左肩
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++right_shoulder object 右肩
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++left_elbow object 左手肘
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++right_elbow object 右手肘
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++left_wrist object 左手腕
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++right_wrist object 右手腕
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++left_hip object 左髋部
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++right_hip object 右髋部
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++left_knee object 左膝
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++right_knee object 右膝
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++left_ankle object 左脚踝
+++x float x坐标
+++y float y坐标
+++score float 概率分数
++right_ankle object 右脚踝
+++x float x坐标
+++y float y坐标
+++score float 概率分数

说明:

1、body_parts,一共21个part,每个part包含x,y两个坐标,如果part被截断,则x、y坐标为part被截断的图片边界位置,part顺序以实际返回顺序为准。

2、接口返回人体坐标框和每个关键点的置信度分数,在应用时可综合置信度score分数,过滤掉置信度低的“无效人体”,建议过滤方法:当关键点得分大于0.2的个数大于3,且人体框的分数大于0.03时,才认为是有效人体。实际应用中,可根据对误识别、漏识别的容忍程度,调整阈值过滤方案,灵活应用。

返回示例

{
	"person_num": 2,
	"person_info": [
		{
			"body_parts": {
				"left_hip": {
					"y": 573,
					"x": 686.09375,
					"score": 0.78743487596512
				},
				"top_head": {
					"y": 242.53125,
					"x": 620,
					"score": 0.87757384777069
				},
				"right_mouth_corner": {
					"y": 308.625,
					"x": 606.78125,
					"score": 0.90121293067932
				},
				"neck": {
					"y": 335.0625,
					"x": 620,
					"score": 0.84662038087845
				},
				"left_shoulder": {
					"y": 361.5,
					"x": 699.3125,
					"score": 0.83550786972046
				},
				"left_knee": {
					"y": 731.625,
					"x": 699.3125,
					"score": 0.83575332164764
				},
				"left_ankle": {
					"y": 877.03125,
					"x": 725.75,
					"score": 0.85220056772232
				},
				"left_mouth_corner": {
					"y": 308.625,
					"x": 633.21875,
					"score": 0.91475087404251
				},
				"right_elbow": {
					"y": 348.28125,
					"x": 461.375,
					"score": 0.81766486167908
				},
				"right_ear": {
					"y": 282.1875,
					"x": 593.5625,
					"score": 0.86551451683044
				},
				"nose": {
					"y": 295.40625,
					"x": 620,
					"score": 0.90894532203674
				},
				"left_eye": {
					"y": 282.1875,
					"x": 633.21875,
					"score": 0.89628517627716
				},
				"right_eye": {
					"y": 282.1875,
					"x": 606.78125,
					"score": 0.89676940441132
				},
				"right_hip": {
					"y": 586.21875,
					"x": 593.5625,
					"score": 0.79803824424744
				},
				"left_wrist": {
					"y": 374.71875,
					"x": 884.375,
					"score": 0.89635348320007
				},
				"left_ear": {
					"y": 295.40625,
					"x": 659.65625,
					"score": 0.86607384681702
				},
				"left_elbow": {
					"y": 361.5,
					"x": 791.84375,
					"score": 0.83910942077637
				},
				"right_shoulder": {
					"y": 348.28125,
					"x": 553.90625,
					"score": 0.85635334253311
				},
				"right_ankle": {
					"y": 890.25,
					"x": 580.34375,
					"score": 0.85149073600769
				},
				"right_knee": {
					"y": 744.84375,
					"x": 580.34375,
					"score": 0.83749794960022
				},
				"right_wrist": {
					"y": 348.28125,
					"x": 368.84375,
					"score": 0.83893859386444
				}
			},
			"location": {
				"height": 703.20654296875,
				"width": 652.61810302734,
				"top": 221.92272949219,
				"score": 0.99269664287567,
				"left": 294.03039550781
			}
		},
		{
			"body_parts": {
				"left_hip": {
					"y": 576,
					"x": 1239.5625,
					"score": 0.84608125686646
				},
				"top_head": {
					"y": 261.15625,
					"x": 1176.59375,
					"score": 0.871442258358
				},
				"right_mouth_corner": {
					"y": 336.71875,
					"x": 1164,
					"score": 0.90951544046402
				},
				"neck": {
					"y": 361.90625,
					"x": 1176.59375,
					"score": 0.85904294252396
				},
				"left_shoulder": {
					"y": 361.90625,
					"x": 1239.5625,
					"score": 0.8512310385704
				},
				"left_knee": {
					"y": 714.53125,
					"x": 1277.34375,
					"score": 0.82312393188477
				},
				"left_ankle": {
					"y": 853.0625,
					"x": 1315.125,
					"score": 0.83786374330521
				},
				"left_mouth_corner": {
					"y": 336.71875,
					"x": 1189.1875,
					"score": 0.90610301494598
				},
				"right_elbow": {
					"y": 387.09375,
					"x": 1025.46875,
					"score": 0.88956367969513
				},
				"right_ear": {
					"y": 311.53125,
					"x": 1138.8125,
					"score": 0.86518502235413
				},
				"nose": {
					"y": 324.125,
					"x": 1176.59375,
					"score": 0.9168484210968
				},
				"left_eye": {
					"y": 311.53125,
					"x": 1189.1875,
					"score": 0.91715461015701
				},
				"right_eye": {
					"y": 311.53125,
					"x": 1164,
					"score": 0.90343600511551
				},
				"right_hip": {
					"y": 576,
					"x": 1164,
					"score": 0.81976848840714
				},
				"left_wrist": {
					"y": 298.9375,
					"x": 1378.09375,
					"score": 0.86095398664474
				},
				"left_ear": {
					"y": 311.53125,
					"x": 1201.78125,
					"score": 0.86899447441101
				},
				"left_elbow": {
					"y": 324.125,
					"x": 1315.125,
					"score": 0.89198768138885
				},
				"right_shoulder": {
					"y": 387.09375,
					"x": 1101.03125,
					"score": 0.85161662101746
				},
				"right_ankle": {
					"y": 878.25,
					"x": 1151.40625,
					"score": 0.83667933940887
				},
				"right_knee": {
					"y": 727.125,
					"x": 1151.40625,
					"score": 0.85485708713531
				},
				"right_wrist": {
					"y": 387.09375,
					"x": 949.90625,
					"score": 0.83042001724243
				}
			},
			"location": {
				"height": 670.80139160156,
				"width": 524.25476074219,
				"top": 241.42504882812,
				"score": 0.98725789785385,
				"left": 902.15216064453
			}
		}
	],
	"log_id": "6362401025381690607"
}
上一篇
调用方式
下一篇
人体检测和属性识别