摘要：
今天，我只使用了CoreML。我这个模型的主要功能是打开相机，然后对准物体。它将自动帮助我们识别相机中的对象，并提供如下近似百分比代码：@IBActionfuncstartClick（_sender:Any）｛startFlag=！StartFlagifstartFlag｛startCaptureVideo（）startButton。setTitle（“stop”，用于：.nor
今天简单使用了下CoreML ， 我的这个模型功能主要是 打开摄像头，然后对准物体，会自动帮我们识别摄像头中的物体，并且给我们大概的百分比值
代码如下：
@IBAction func startClick(_ sender: Any) {
startFlag = !startFlag
if startFlag {
startCaptureVideo()
startButton.setTitle("stop", for: .normal)
}else {
startButton.setTitle("start", for: .normal)
stop()
}
}
//1、初始化CaptureSeason
lazy var captureSession:AVCaptureSession? = {
//1.1 实例化
let captureSession = AVCaptureSession()
captureSession.sessionPreset = .photo
//1.2 获取默认设备
guard let captureDevice = AVCaptureDevice.default(for: .video) else {
return nil
}
guard let captureDeviceInput = try? AVCaptureDeviceInput(device: captureDevice) else {
return nil
}
captureSession.addInput(captureDeviceInput)
let previewLayer = AVCaptureVideoPreviewLayer(session: captureSession)
view.layer.addSublayer(previewLayer)
previewLayer.frame = view.frame
let dataOutput = AVCaptureVideoDataOutput()
dataOutput.setSampleBufferDelegate(self, queue: DispatchQueue(label: "kingboVision"))
captureSession.addOutput(dataOutput)
return captureSession
}()
func startCaptureVideo(){
captureSession?.startRunning()
}
func stop() {
captureSession?.stopRunning()
}
func captureOutput(_ output: AVCaptureOutput, didOutput sampleBuffer: CMSampleBuffer, from connection: AVCaptureConnection) {
print("Camera was safe ")
guard let pixelBuffer:CVPixelBuffer = CMSampleBufferGetImageBuffer(sampleBuffer) else
{
print("nil object ")
return
}
guard let model = try? VNCoreMLModel(for: SqueezeNet().model) else {return}
let request = VNCoreMLRequest(model: model) { [unowned self](finishRequest, error) in
//
guard let results = finishRequest.results as? [VNClassificationObservation] else {
return
}
guard let first = results.first else {
return
}
DispatchQueue.main.async {
self.descriptionLabel.text = "(first.identifier): (first.confidence * 100)%,"
}
}
print("start to VN")
try? VNImageRequestHandler(cvPixelBuffer: pixelBuffer, options: [:]).perform([request])
}
分为以下几步操作
1、我们工程inf.plist 文件中添加摄像头权限说明
Privacy - Camera Usage Description 相机权限设置 Privacy - Microphone Usage Description 麦克风权限设置
2、头文件引用
import AVKit import Vision
3、准备模型
模型准备我这里是使用App 官网上的SqueezeNet模型
4、将模型拖动到工程里面即可
5、代码编写部分
5.1 初始化AVCaptureSession
let captureSession = AVCaptureSession()
captureSession.sessionPreset = .photo
5.2 获取当前iPhone 设备控制，当作输入数据来源
guard let captureDevice = AVCaptureDevice.default(for: .video) else {
return nil
}captureSession.addInput(captureDeviceInput)
5.3 设置输出数据
//设置输出数据处理对象
let dataOutput = AVCaptureVideoDataOutput() dataOutput.setSampleBufferDelegate(self, queue: DispatchQueue(label: "kingboVision")) captureSession.addOutput(dataOutput)
AVCaptureVideoDataOutput 设置代理，处理数据流图片数据，因此当前ViewController需要实现AVCaptureVideoDataOutputSampleBufferDelegate 的代理，如下
class ViewController: UIViewController ,AVCaptureVideoDataOutputSampleBufferDelegate{
......
func captureOutput(_ output: AVCaptureOutput, didOutput sampleBuffer: CMSampleBuffer, from connection: AVCaptureConnection) {
//将视频数据转成CVPixBuffer
guard let pixelBuffer:CVPixelBuffer = CMSampleBufferGetImageBuffer(sampleBuffer) else
{
print("nil object ")
return
}
// 这里SquzzedNet是我从苹果中直接获取的模型
guard let model = try? VNCoreMLModel(for: SqueezeNet().model) else {return}
let request = VNCoreMLRequest(model: model) { [unowned self](finishRequest, error) in
//处理数据分析进行业务处理
guard let results = finishRequest.results as? [VNClassificationObservation] else {
return
}
guard let first = results.first else {
return
}
DispatchQueue.main.async {
self.descriptionLabel.text = "(first.identifier): (first.confidence * 100)%,"
}
}
print("start to VN")
try? VNImageRequestHandler(cvPixelBuffer: pixelBuffer, options: [:]).perform([request])
}
.....
}
5.4 将当前的摄像头显示在指定的View 上，（我们这里使用当前VC 的view)
let previewLayer = AVCaptureVideoPreviewLayer(session: captureSession)
view.layer.addSublayer(previewLayer)
previewLayer.frame = view.frame
效果如下：
