iOS 人脸检测
2018-02-01 本文已影响529人
li_礼光
如果用iOS 原生的CIDetector来做人脸检测
/** CIDetector发现的脸部特征。
所有的位置都是相对于原始图像. */
NS_CLASS_AVAILABLE(10_7, 5_0)
@interface CIFaceFeature : CIFeature
{
CGRect bounds;
BOOL hasLeftEyePosition;
CGPoint leftEyePosition;
BOOL hasRightEyePosition;
CGPoint rightEyePosition;
BOOL hasMouthPosition;
CGPoint mouthPosition;
BOOL hasTrackingID;
int trackingID;
BOOL hasTrackingFrameCount;
int trackingFrameCount;
BOOL hasFaceAngle;
float faceAngle;
BOOL hasSmile;
BOOL leftEyeClosed;
BOOL rightEyeClosed;
}
/**脸部各个基点的坐标。
请注意，左眼是脸左侧的眼睛
从观察者的角度来看。 这不是左眼
主体的视角. */
@property (readonly, assign) CGRect bounds; //指示图像坐标中的人脸位置和尺寸的矩形。
@property (readonly, assign) BOOL hasLeftEyePosition; //指示检测器是否找到了人脸的左眼。
@property (readonly, assign) CGPoint leftEyePosition; //左眼的坐标
@property (readonly, assign) BOOL hasRightEyePosition; //指示检测器是否找到了人脸的右眼。
@property (readonly, assign) CGPoint rightEyePosition; //右眼的坐标
@property (readonly, assign) BOOL hasMouthPosition; //指示检测器是否找到了人脸的口
@property (readonly, assign) CGPoint mouthPosition; //口的坐标
@property (readonly, assign) BOOL hasTrackingID; //指示面部对象是否具有跟踪ID。
@property (readonly, assign) int trackingID;
@property (readonly, assign) BOOL hasTrackingFrameCount; //指示面部对象的布尔值具有跟踪帧计数。
@property (readonly, assign) int trackingFrameCount; //跟踪帧计数
@property (readonly, assign) BOOL hasFaceAngle; //指示是否有关于脸部旋转的信息可用。
@property (readonly, assign) float faceAngle; //旋转是以度数逆时针测量的，其中零指示在眼睛之间画出的线相对于图像方向是水平的。
@property (readonly, assign) BOOL hasSmile; //是否有笑脸
@property (readonly, assign) BOOL leftEyeClosed; //左眼是否闭上
@property (readonly, assign) BOOL rightEyeClosed; //右眼是否闭上
@end
关于trackingID
核心图像提供了在视频流中检测到的脸部的跟踪标识符，您可以使用该标识符来识别在一个视频帧中检测到的CIFaceFeature对象是在先前视频帧中检测到的同一个脸部。
只有在框架中存在人脸并且不与特定人脸相关联时，该标识符才会一直存在。
换句话说，如果脸部移出视频帧并在稍后返回到帧中，则分配另一个ID。 （核心图像检测面部，但不识别特定的面部。）
这个有点抽象,暂时不理解
测试
打印数据
bounds : {{237.5, 115}, {435, 435}}
hasLeftEyePosition : 1
leftEyePosition : {381.25, 405}
hasRightEyePosition : 1
rightEyePosition : {550, 388.75}
hasMouthPosition : 1
mouthPosition : {450, 225}
hasTrackingID : 0
trackingID : 0
hasTrackingFrameCount : 0
trackingFrameCount : 0
hasFaceAngle : 1
faceAngle : 3.000000rightEyeClosed
hasSmile : 0
leftEyeClosed : 0
rightEyeClosed : 0
发现
hasSmile, leftEyeClosed, rightEyeClosed 检测都不准确
整体来说除了:
bounds ,leftEyePosition, rightEyePosition, mouthPosition这四个属性有可用价值,其他的可用价值体验就不那么明显了
也就是
可以用,但不推荐
实现思路
- 正常创建一个相机,初始化相机的所有功能(怎么实现iOS相机功能,请教度娘)
- (预览的时候)获取相机的视频帧数据,也就是在相机预览的时候,会不断的调用以下的方法获取帧数据
(遵循协议 : AVCaptureVideoDataOutputSampleBufferDelegate)
- (void)captureOutput:(AVCaptureOutput *)captureOutput
didOutputSampleBuffer:(CMSampleBufferRef)sampleBuffer
fromConnection:(AVCaptureConnection *)connection;
- 通过得到的sampleBuffer 帧缓冲数据来拿到UIImage对象 CIImage -> CGImageRef -> UIImage
- 对每一帧的UIImage做人脸检测
方法:
// 通过抽样缓存数据创建一个UIImage对象
- (UIImage *)imageFromSampleBuffer:(CMSampleBufferRef) sampleBuffer
{
//CIImage -> CGImageRef -> UIImage
CVImageBufferRef imageBuffer = CMSampleBufferGetImageBuffer(sampleBuffer); //拿到缓冲区帧数据
CIImage *ciImage = [CIImage imageWithCVPixelBuffer:imageBuffer]; //创建CIImage对象
CIContext *temporaryContext = [CIContext contextWithOptions:nil]; //创建上下文
CGImageRef cgImageRef = [temporaryContext createCGImage:ciImage fromRect:CGRectMake(0, 0, CVPixelBufferGetWidth(imageBuffer), CVPixelBufferGetHeight(imageBuffer))];
UIImage *result = [[UIImage alloc] initWithCGImage:cgImageRef scale:1.0 orientation:UIImageOrientationLeftMirrored]; //创建UIImage对象
CGImageRelease(cgImageRef); //释放上下文
return result;
}
/**识别脸部*/
-(NSArray *)detectFaceWithImage:(UIImage *)faceImag
{
//此处是CIDetectorAccuracyHigh，若用于real-time的人脸检测，则用CIDetectorAccuracyLow，更快
CIDetector *faceDetector = [CIDetector detectorOfType:CIDetectorTypeFace
context:nil
options:@{CIDetectorAccuracy: CIDetectorAccuracyHigh}];
CIImage *ciimg = [CIImage imageWithCGImage:faceImag.CGImage];
NSArray *features = [faceDetector featuresInImage:ciimg];
CIFaceFeature *faceFeature = [features firstObject];
if (faceFeature) {
NSLog(@"bounds : %@",NSStringFromCGRect(faceFeature.bounds));
NSLog(@"hasLeftEyePosition : %d",faceFeature.hasLeftEyePosition);
NSLog(@"leftEyePosition : %@",NSStringFromCGPoint(faceFeature.leftEyePosition));
NSLog(@"hasRightEyePosition : %d",faceFeature.hasRightEyePosition);
NSLog(@"rightEyePosition : %@",NSStringFromCGPoint(faceFeature.rightEyePosition));
NSLog(@"hasMouthPosition : %d",faceFeature.hasMouthPosition);
NSLog(@"mouthPosition : %@",NSStringFromCGPoint(faceFeature.mouthPosition));
NSLog(@"hasTrackingID : %d",faceFeature.hasTrackingID);
NSLog(@"trackingID : %d",faceFeature.trackingID);
NSLog(@"hasTrackingFrameCount : %d",faceFeature.hasTrackingFrameCount);
NSLog(@"trackingFrameCount : %d",faceFeature.trackingFrameCount);
NSLog(@"hasFaceAngle : %d",faceFeature.hasFaceAngle);
NSLog(@"faceAngle : %frightEyeClosed",faceFeature.faceAngle);
NSLog(@"hasSmile : %d",faceFeature.hasSmile);
NSLog(@"leftEyeClosed : %d",faceFeature.leftEyeClosed);
NSLog(@"rightEyeClosed : %d",faceFeature.rightEyeClosed);
NSLog(@"\n\n\n");
}
return features;
}
关于人脸识别的一些思考 :
- 人脸识别/检测的应用区分场景,
比如 : 美图/视频编辑, 社区监控,人脸门禁,智能家居,广告展示,老人购物等… - 美图/视频编辑 : ==> 图片数据在客户端上处理.
- 社区监控,人脸门禁,智能家居,广告展示,老人购物 : ==> 图片数据在服务器上处理.
