Published February 10, 2019 © MIT

Auto Focus of Detected Object From a Live Stream

In this project objects like cars and person get detected through a live video stream using a webcam and the background is made blurred.

IntermediateFull instructions provided4 hours1,437

Auto Focus of Detected Object From a Live Stream

Things used in this project

Hardware components

Raspberry Pi 3 Model B

logitech webcam

Software apps and online services

OpenCV

Story

About Project

This project uses a normal webcam to capture video stream and detect objects like cars and people and blurs the background whenever it encounters those objects.

Usage

As soon as an object of interest is detected, background is made blurred and focus is on the object itself. This kind of logic can be used to easily detect objects of interests in video surveillance by eliminating detailed and noisy backgrounds.

Car in Focus with blurred background in a live video stream

Person as object to be focused on

There is another logic which blurs the object of interest without manipulating the background. In this variant a particular scene can be detected an blurred during live video capture.

Object blur without affecting background

How It Was Done

I have used Intel Movidius™ on a Raspberry Pi B+ board using a pretrained Caffe model to detect objects. After detecting the object of interest, I used Gaussian Blur method to blur the background and alpha blended with the detected object.

Code

  # get cordinates of bounding box
        (y1, x1) = output_dict.get('detection_boxes_' + str(i))[0]
        (y2, x2) = output_dict.get('detection_boxes_' + str(i))[1]

        if detection == True:
          height=y2-y1
          width=x2-x1
          mask = np.zeros((480,640),np.uint8)
          cv2.rectangle(mask,(x1,y1),(x1+width,y1+height),(255,255,255),-1)
          mask = cv2.GaussianBlur(mask, (37,37), 0 )
          blurred = cv2.GaussianBlur(frame, (37,37), 0) 
          alpha = cv2.cvtColor(255-mask, cv2.COLOR_GRAY2BGR)/255.0
          frame = cv2.convertScaleAbs(frame*(1-alpha) + blurred*alpha)
          detection=False
       
    # If a display is available, show the image on which inference was performed
    if 'DISPLAY' in os.environ:
        cv2.imshow( 'Auto Focus', frame )

for i in range( 0, output_dict['num_detections'] ):
        if( (labels[ int(output_dict['detection_classes_' + str(i)]) ]=="15: person") or  (labels[ int(output_dict['detection_classes_' + str(i)]) ]=="7: car" ) ):
           detection=True

        # get cordinates of bounding box
        (y1, x1) = output_dict.get('detection_boxes_' + str(i))[0]
        (y2, x2) = output_dict.get('detection_boxes_' + str(i))[1]

        if detection == True:
          height=y2-y1
          width=x2-x1
          mask = np.zeros((480,640),np.uint8)
          cv2.rectangle(mask,(x1,y1),(x1+width,y1+height),(255,255,255),-1)
          mask = cv2.GaussianBlur(mask, (37,37), 0 )
          blurred = cv2.GaussianBlur(frame, (37,37), 0) 
          alpha = cv2.cvtColor(mask, cv2.COLOR_GRAY2BGR)/255.0
          frame = cv2.convertScaleAbs(frame*(1-alpha) + blurred*alpha)
          detection=False

Credits

Prateek Pant

3 projects • 6 followers

Video Enthusiast

Auto Focus of Detected Object From a Live Stream

Things used in this project

Hardware components

Software apps and online services

Story

About Project

Usage

How It Was Done

Schematics

setup

Code

Background Blur

Object blur

Credits

Prateek Pant

Comments

Embed the widget on your own site

Auto Focus of Detected Object From a Live Stream

Auto Focus of Detected Object From a Live Stream

Things used in this project

Hardware components

Software apps and online services

Story

About Project

Usage

How It Was Done

Schematics

setup

Code

Background Blur

Object blur

Credits

Prateek Pant

Comments

Related channels and tags