本文将讲到，如何利用百度ai的人体分析sdk在unity中实现摄像头实时画面的手势识别

工程链接在文末

（本人使用的是unity，unity以上都可以，不然.NET版本不支持）

1.准备工作：首先在百度AI的官网，下载人体分析C# SDK

然后登陆控制台，新建一个人体分析的应用

然后你会获取到该应用的API_KEY和SECRET_KEY，后续开发需要使用

2.准备工作做完后，新建一个unity工程，在Asset下新建一个文件夹取名Plugins，将第一步下载的SDK中两个DLL文件复制进来

然后就可以编程了

代码很简单，注释也写的很清楚，我只讲下思路：

其实百度AI的SDK是只单纯的识别图片，而我们要做的识别摄像头实时画面中的手势，其实方法很简单

此时我们只需在固定时间截图摄像头画面就可以了，本文设定每两秒截一次图，然后调用SDK识别此图中的手势，就能实现我们所需要的实时画面中的手势

代码如下：

using System.Collections;using System.Collections.Generic;using UnityEngine;using Baidu.Aip.BodyAnalysis;using System.IO;using UnityEngine.UI;public class FaceDetect : MonoBehaviour{public string app_id;public string api_key;public string secret_key;Body client;private string deviceName;private WebCamTexture webTex;//百度AI返回的结果数据public Text resultMsg;//提取其中的手势名称public Text detectedGestureMsg;void Awake(){.ServicePointManager.ServerCertificateValidationCallback +=delegate (object sender, System.Security.Cryptography.X509Certificates.X509Certificate certificate,System.Security.Cryptography.X509Certificates.X509Chain chain,.Security.SslPolicyErrors sslPolicyErrors){return true; // **** Always accept};}// Use this for initializationvoid Start(){api_key = "你自己的API_KEY";secret_key = "你自己的SECRET_KEY";StartCoroutine(CallCamera());client = new Body(api_key, secret_key);client.Timeout = 60000; // 修改超时时间}// Update is called once per framevoid Update(){CaptureScreen();}IEnumerator CallCamera(){yield return Application.RequestUserAuthorization(UserAuthorization.WebCam);if (Application.HasUserAuthorization(UserAuthorization.WebCam)){WebCamDevice[] devices = WebCamTexture.devices;deviceName = devices[0].name;//设置摄像机摄像的区域webTex = new WebCamTexture(deviceName, 1024, 768, 20);webTex.Play();//开始摄像transform.GetComponent<RawImage>().texture = webTex;}}public float timer = 0;//截屏void CaptureScreen(){timer += Time.deltaTime;//每隔两秒检测一次if (timer > 2){//删除上一次检测的图片File.Delete(Application.streamingAssetsPath + "/capture.jpg");CapturePhoto();timer = 0;}}public int width;public int height;//截图摄像头public Camera cameras;public string fileName;public void CapturePhoto(){Texture2D screenShot;RenderTexture rt = new RenderTexture(width, height, 1);cameras.targetTexture = rt;cameras.Render();RenderTexture.active = rt;screenShot = new Texture2D(width, height, TextureFormat.RGB24, false);screenShot.ReadPixels(new Rect(0, 0, width, height), 0, 0);screenShot.Apply();//运行此行代码前，先手动在Asset路径下新建一个StreamingAsset文件夹fileName = Application.streamingAssetsPath + "/capture.jpg";// byte[] bytes = screenShot.EncodeToJPG();ScaleTextureCutOut(screenShot, 0, 0, 1024, 768);Debug.Log(string.Format("截屏了一张照片: {0}", fileName));}//切图byte[] ScaleTextureCutOut(Texture2D originalTexture, int pos_x, int pos_y, float originalWidth, float originalHeight){Color[] pixels = new Color[(int)(originalWidth * originalHeight)];//要返回的新图Texture2D newTexture = new Texture2D(Mathf.CeilToInt(originalWidth), Mathf.CeilToInt(originalHeight));//批量获取点像素pixels = originalTexture.GetPixels(pos_x, pos_y, (int)originalWidth, (int)originalHeight);newTexture.SetPixels(pixels);newTexture.anisoLevel = 2;newTexture.Apply();//这一步把裁剪的新图片存下来byte[] jpgData = newTexture.EncodeToJPG();System.IO.File.WriteAllBytes(fileName, jpgData);GestureDemo(fileName);return jpgData;}public void GestureDemo(string filesPath){var image = File.ReadAllBytes(filesPath);try{var result = client.Gesture(image);resultMsg.text = result.ToString();string[] msgArr = resultMsg.text.Split(',');//单独提取classnamefor (int i = 0; i < msgArr.Length; i++){if (msgArr[i].Contains("classname")){string[] strArr = msgArr[i].Split(':');detectedGestureMsg.text = strArr[1];break;}}}catch (System.Exception){throw;}}}

工程结构如下：

运行前记得填写自己的API_KEY和SECRET_KEY

工程链接：

链接：/s/1CbGjzRHHVnt5UZ2rrGdCdQ

提取码：nu3y