Skip to content

Instantly share code, notes, and snippets.

@jsbain

jsbain/detector.py

Created Jun 11, 2020
Embed
What would you like to do?
detector.py
# coding: utf-8
# based on Cethric's image capture gist....
FRAME_PROC_INTERVAL=15 #num frames to skip. 1=go as fast as possible, 5=every fifth frame
import ui
from objc_util import *
import ctypes
from objc_util import autoreleasepool
AVCaptureDevice = ObjCClass('AVCaptureDevice')
AVCaptureDeviceInput = ObjCClass('AVCaptureDeviceInput')
AVCaptureVideoDataOutput = ObjCClass('AVCaptureVideoDataOutput')
AVCaptureSession = ObjCClass('AVCaptureSession')
AVCaptureVideoPreviewLayer = ObjCClass('AVCaptureVideoPreviewLayer')
AVCaptureStillImageOutput = ObjCClass('AVCaptureStillImageOutput')
AVCaptureConnection = ObjCClass('AVCaptureConnection')
UIImage = ObjCClass('UIImage')
CIImage = ObjCClass('CIImage')
CIDetector = ObjCClass('CIDetector')
CIDetectorTypeRectangle=ObjCInstance(c_void_p.in_dll(c,'CIDetectorTypeRectangle'))
CIDetectorTypeFace=ObjCInstance(c_void_p.in_dll(c,'CIDetectorTypeFace'))
kCGBitmapAlphaInfoMask = 0x1F
kCGBitmapFloatComponents = 1 << 8
kCGBitmapByteOrderMask = 0x7000
kCGBitmapByteOrderDefault = 0 << 12
kCGBitmapByteOrder16Little = 1 << 12
kCGBitmapByteOrder32Little = 2 << 12
kCGBitmapByteOrder16Big = 3 << 12
kCGBitmapByteOrder32Big = 4 << 12
kCGImageAlphaNone = 0
kCGImageAlphaPremultipliedLast = 1
kCGImageAlphaPremultipliedFirst = 2
kCGImageAlphaLast = 3
kCGImageAlphaFirst = 4
kCGImageAlphaNoneSkipLast = 5
kCGImageAlphaNoneSkipFirst = 6
kCGImageAlphaOnly = 7
kDroppedFrameReason = ObjCInstance(c_void_p.in_dll(c,'kCMSampleBufferAttachmentKey_DroppedFrameReason'))
CMGetAttachment = c.CMGetAttachment
CMGetAttachment.argtypes = [c_void_p, c_void_p, c_void_p]
CMGetAttachment.restype = c_void_p
class CMTime(ctypes.Structure):
_fields_ = [
('CMTimeValue', ctypes.c_int64),
('CMTimeScale', ctypes.c_int32),
('CMTimeEpoch', ctypes.c_int64),
('CMTimeFlags', ctypes.c_uint32),
]
def CMTimeMake(value, scale):
cm = CMTime()
cm.CMTimeScale = scale
cm.CMTimeValue = value
return cm
def dispatch_queue_create(name, parent):
func = c.dispatch_queue_create
func.argtypes = [ctypes.c_char_p, ctypes.c_void_p]
func.restype = ctypes.c_void_p
return ObjCInstance(func(name, parent))
def dispatch_get_current_queue():
func=c.dispatch_get_current_queue
func.argtypes=[]
func.restype=c_void_p
return ObjCInstance(func())
def dispatch_release(queue_obj):
#raise RuntimeError('This is not the method you are looking for')
func = c.dispatch_release
func.argtyps = [ctypes.c_void_p]
func.restype = None
return func(ObjCInstance(queue_obj).ptr)
def CMSampleBufferGetImageBuffer(buffer):
func = c.CMSampleBufferGetImageBuffer
func.argtypes = [ctypes.c_void_p]
func.restype = ctypes.c_void_p
return ObjCInstance(func(buffer))
def CVPixelBufferLockBaseAddress(imagebuffer, param_0):
func = c.CVPixelBufferLockBaseAddress
func.argtypes = [ctypes.c_void_p, ctypes.c_int]
func.restype = None
return func(ObjCInstance(imagebuffer).ptr, param_0)
def CVPixelBufferUnlockBaseAddress(imagebuffer, param_0):
func = c.CVPixelBufferUnlockBaseAddress
func.argtypes = [ctypes.c_void_p, ctypes.c_int]
func.restype = None
return func(ObjCInstance(imagebuffer).ptr, param_0)
def CVPixelBufferGetBaseAddress(imagebuffer):
func = c.CVPixelBufferGetBaseAddress
func.argtypes = [ctypes.c_void_p]
func.restype = ctypes.c_void_p
return func(ObjCInstance(imagebuffer).ptr)
def CVPixelBufferGetBytesPerRow(imagebuffer):
func = c.CVPixelBufferGetBytesPerRow
func.argtypes = [ctypes.c_void_p]
func.restype = ctypes.c_int
return func(ObjCInstance(imagebuffer).ptr)
def CVPixelBufferGetWidth(imagebuffer):
func = c.CVPixelBufferGetWidth
func.argtypes = [ctypes.c_void_p]
func.restype = ctypes.c_int
return func(ObjCInstance(imagebuffer).ptr)
def CVPixelBufferGetHeight(imagebuffer):
func = c.CVPixelBufferGetHeight
func.argtypes = [ctypes.c_void_p]
func.restype = ctypes.c_int
return func(ObjCInstance(imagebuffer).ptr)
def CGColorSpaceCreateDeviceRGB():
func = c.CGColorSpaceCreateDeviceRGB
func.argtypes = None
func.restype = ctypes.c_void_p
return ObjCInstance(func())
def CGBitmapContextCreate(baseAddress, width, height, param_0, bytesPerRow, colorSpace, flags):
func = c.CGBitmapContextCreate
func.argtypes = [ctypes.c_void_p, ctypes.c_int, ctypes.c_int, ctypes.c_int, ctypes.c_int, ctypes.c_void_p, ctypes.c_int32]
func.restype = ctypes.c_void_p
result = func(baseAddress, int(width), int(height), int(param_0), int(bytesPerRow), ObjCInstance(colorSpace).ptr, flags)
if result is not None:
return ObjCInstance(result)
else:
raise RuntimeError('Failed to create context')
def CGContextRelease(context):
func = c.CGContextRelease
func.argtypes = [ctypes.c_void_p]
func.restype = None
return func(ObjCInstance(context).ptr)
def CGColorSpaceRelease(colorSpace):
func = c.CGColorSpaceRelease
func.argtypes = [ctypes.c_void_p]
func.restype = None
return func(ObjCInstance(colorSpace))
def CGBitmapContextCreateImage(context):
func = c.CGBitmapContextCreateImage
func.argtypes = [ctypes.c_void_p]
func.restype = ctypes.c_void_p
return ObjCInstance(func(ObjCInstance(context).ptr))
def CGImageRelease(image):
func = c.CGImageRelease
func.argtypes = [ctypes.c_void_p]
func.restype = None
return func(ObjCInstance(image).ptr)
def UIImageFromSampleBuffer(buffer):
imagebuffer = CMSampleBufferGetImageBuffer(buffer)
CVPixelBufferLockBaseAddress(imagebuffer, 0)
baseAddress = CVPixelBufferGetBaseAddress(imagebuffer)
width = CVPixelBufferGetWidth(imagebuffer)
height = CVPixelBufferGetHeight(imagebuffer)
colorSpace = CGColorSpaceCreateDeviceRGB()
flags = kCGBitmapByteOrderDefault | kCGImageAlphaNoneSkipFirst
context = CGBitmapContextCreate(baseAddress, width / 4, height / 4, 8, 4 * width, colorSpace, flags)
quartzImage = CGBitmapContextCreateImage(context)
CVPixelBufferUnlockBaseAddress(imagebuffer, 0)
CGContextRelease(context)
CGColorSpaceRelease(colorSpace)
image = UIImage.imageWithCGImage_(quartzImage)
CGImageRelease(quartzImage)
return image
def UIImagePNGRepresentation(image):
func = c.UIImagePNGRepresentation
func.argtypes = [ctypes.c_void_p]
func.restype = ctypes.c_void_p
return ObjCInstance(func(ObjCInstance(image).ptr))
def UIImageJPEGRepresentation(image, scale=32):
func = c.UIImageJPEGRepresentation
func.argtypes = [ctypes.c_void_p, ctypes.c_float]
func.restype = ctypes.c_void_p
return ObjCInstance(func(ObjCInstance(image).ptr, scale))
def detect_faces(samplebuffer):
imagebuffer = CMSampleBufferGetImageBuffer(samplebuffer)
CVPixelBufferLockBaseAddress(imagebuffer, 0)
baseAddress = CVPixelBufferGetBaseAddress(imagebuffer)
width = CVPixelBufferGetWidth(imagebuffer)
height = CVPixelBufferGetHeight(imagebuffer)
ciimage = CIImage.imageWithCVPixelBuffer_((imagebuffer))
options = {'CIDetectorAccuracy': 'CIDetectorAccuracyHigh'}
face_detector = CIDetector.detectorOfType_context_options_('CIDetectorTypeFace', None, options)
rectangle_detector = CIDetector.detectorOfType_context_options_('CIDetectorTypeRectangle', None, options)
faces = face_detector.featuresInImage_(ciimage)
#if faces:
# faces=list(ObjCInstance(faces))
#else:
faces=[]
rects = rectangle_detector.featuresInImage_(ciimage)
if rects:
rects=list(ObjCInstance(rects))
else:
rects=[]
CVPixelBufferUnlockBaseAddress(imagebuffer, 0)
return faces, rects, (width, height)
view = None
@on_main_thread
def change_image(image):
if view is not None:
ObjCInstance(view['imageview1']).setImage_(ObjCInstance(image))
# view['imageview1'].image = ui.Image.named('test.png')
@on_main_thread
def set_label(txt):
if view is not None:
view['lbl'].text=txt
import time
@on_main_thread
def compute_fps():
a=.1 # ewma coefficient to smooth fps display
if view is not None:
t=time.perf_counter()
view.fps=a/(t-view.last_time)+(1-a)*view.fps
view.last_time=t
faces=getattr(view,'faces',[])
rects=getattr(view, 'rects',[])
view.heartbeat=(view.heartbeat+1)%2
error_str=''
set_label(
'fps={:0.1f} proc:{:9d} faces:{} rects:{} {} {}'.format(view.fps,view.processed_frames, len(faces),len(rects),view.heartbeat, error_str))
def captureOutput_didOutputSampleBuffer_fromConnection_(_cmd, _self, _output, _buffer, _connection):
view.frame_count=(view.frame_count+1)
if view.frame_count>=FRAME_PROC_INTERVAL:
view.frame_count=0
buffer = ObjCInstance(_buffer)
#image = UIImageFromSampleBuffer(buffer)
faces,rects, image_size=detect_faces(buffer)
view.faces=faces
view.rects=rects
view.image_size=image_size
view.processed_frames=view.processed_frames+1
#imageRep = UIImagePNGRepresentation(image)
#imageRep = UIImageJPEGRepresentation(image)
#imageRep.writeToFile_atomically_('test.png', True)
#imageRep.writeToFile_atomically_('test.jpg', True)
#change_image(image)
compute_fps()
pv.set_needs_display()
def captureOutput_didDropSampleBuffer_fromConnection_(_self,_cmd,output, sample_buffer, conn):
reason=CMGetAttachment(sample_buffer, kDroppedFrameReason, None)
view.frame_count=(view.frame_count+1)
if reason:
faces=getattr(view,'faces',[])
rects=getattr(view, 'rects',[])
error_str=str(ObjCInstance(reason))
set_label(
'fps={:0.1f} proc:{:9d} faces:{} rects:{} {} {}'.format(view.fps,view.processed_frames, len(faces),len(rects),view.heartbeat, error_str))
delegate_call = create_objc_class('delegate_call', protocols=['AVCaptureVideoDataOutputSampleBufferDelegate'], methods=[captureOutput_didOutputSampleBuffer_fromConnection_,
captureOutput_didDropSampleBuffer_fromConnection_])
DESIRED_FPS = 5
@on_main_thread
def set_frame_rate(inputDevice, captureSession, desired_fps):
#this doesnt seem to work at all!
supported_rates=inputDevice.activeFormat().videoSupportedFrameRateRanges()[0]
#print('min support',supported_rates.minFrameRate())
if (desired_fps) <= supported_rates.maxFrameRate() and desired_fps >= supported_rates.minFrameRate():
desired_intervalnum=600//desired_fps
err_ptr = c_void_p()
captureSession.beginConfiguration()
inputDevice.lockForConfiguration_(byref(err_ptr))
if err_ptr:
raise()
inputDevice.setActiveVideoMinFrameDuration_((desired_intervalnum,600,0,0))
#print('a,b',inputDevice.activeVideoMinFrameDuration().a ,inputDevice.activeVideoMinFrameDuration().b)
inputDevice.setActiveVideoMinFrameDuration_((desired_intervalnum,600,0,0))
inputDevice.unlockForConfiguration()
captureSession.commitConfiguration()
#print('a,b',inputDevice.activeVideoMinFrameDuration().a ,inputDevice.activeVideoMinFrameDuration().b)
class CameraView(ui.View):
@on_main_thread
def __init__(self, *args, **kwargs):
ui.View.__init__(self, *args, **kwargs)
self.inputDevice = AVCaptureDevice.devices()[0]
#(CMTimeMake(2,1), argtypes=[CMTime], restype=None)
self.captureInput = AVCaptureDeviceInput.deviceInputWithDevice_error_(self.inputDevice, None)
if not self.captureInput:
print('Failed to create device')
exit()
self.captureOutput = AVCaptureVideoDataOutput.alloc().init()
self.captureSession = AVCaptureSession.alloc().init()
self.captureSession.setSessionPreset_('AVCaptureSessionPresetHigh')
if self.captureSession.canAddInput_(self.captureInput):
self.captureSession.addInput_(self.captureInput)
if self.captureSession.canAddOutput_(self.captureOutput):
self.captureSession.addOutput_(self.captureOutput)
self.captureVideoPreviewLayer = AVCaptureVideoPreviewLayer.layerWithSession_(self.captureSession)
queue_test = dispatch_queue_create(b'imageDispatch', None)
#queue_test = dispatch_get_current_queue()
#self.inputDevice.lockForConfiguration_(None)
print(self.captureOutput.connections()[0].videoMinFrameDuration().a)
callback = delegate_call.alloc().init()
self.captureOutput.setSampleBufferDelegate_queue_(callback, queue_test)
self.queue=queue_test
def present(self, *args, **kwargs):
ui.View.present(self, *args, **kwargs)
self.set_layer()
@on_main_thread
def set_layer(self):
v = ObjCInstance(self)
self.captureVideoPreviewLayer.setFrame_(v.bounds())
self.captureVideoPreviewLayer.setVideoGravity_('AVLayerVideoGravityResizeAspectFill')
v.layer().addSublayer_(self.captureVideoPreviewLayer)
set_frame_rate(self.inputDevice,self.captureSession, DESIRED_FPS)
self.captureSession.startRunning()
@on_main_thread
def will_close(self):
self.captureSession.stopRunning()
dispatch_release(self.queue)
class PathView(ui.View):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
def draw_rects(self, rects, color):
ui.set_color(color)
for rect in rects:
rect_bounds = rect.bounds()
# View は X軸=375 Y軸=667
# 画像のX軸Y軸をViewのY軸X軸に対応させ、サイズを調整
x = rect_bounds.origin.y * self.height / view.image_size[0]
y = rect_bounds.origin.x * self.width / view.image_size[1]
w = rect_bounds.size.height * self.height / view.image_size[0]
h = rect_bounds.size.width * self.width / view.image_size[1]
print(x,y,w,h)
path = ui.Path.rect(x, y, w * 1.3, h)
path.fill()
path=ui.Path.rect(100,100,200,200).stroke()
def draw(self):
self.draw_rects(view.rects, (1,0,0,0.5))
self.draw_rects(view.faces, (0,1,0,0.5))
class CustomView(ui.View):
def __init__(self, *args, **kwargs):
ui.View.__init__(self, *args, **kwargs)
self.frame = (0,0, 800,600)
self.fps=0
self.heartbeat=0
self.frame_count=0
self.processed_frames=0
self.last_time=time.perf_counter()
self.faces=[]
self.rects=[]
def did_load(self):
self['camera'].set_layer()
#self['imageview1'].image = ui.Image.named('test:Numbers')
def present(self, *args, **kwargs):
ui.View.present(self, *args, **kwargs)
def will_close(self):
self['camera'].will_close()
view=CustomView()
cv=CameraView(frame=(0,0,400,600), name='camera')
pv=PathView(frame=(0,0,400,600), name='path')
#iv=ui.ImageView(name='imageview1', frame=(400,0,400,600))
lbl=ui.Label(name='lbl',frame=(0,0,800,200))
lbl.text_color='#fbff99'
view.add_subview(cv)
view.add_subview(pv)
#view.add_subview(iv)
view.add_subview(lbl)
view.did_load()
view.present()
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment