Face tracking using Xailient Face Detector and dlib on a video file

Xailient_dlib_facetracker_videofile.py

import sys
import dlib
import cv2
from picamera.array import PiRGBArray
from picamera import PiCamera
import time
import imutils
import numpy as np
from operator import itemgetter
import math
import dnn

dnn.FaceDetector.set_param("resolution", dnn.RESOLUTION.HIGH)
detectum = dnn.FaceDetector()
THRESHOLD = 0.8 # Value between 0 and 1 for confidence score

# initialize the camera and grab a reference to the raw camera capture
RES_W = 640 # 1280 # 640 # 256 # 320 # 480 # pixels
RES_H = 480 # 720 # 480 # 144 # 240 # 360 # pixels

# Create a VideoCapture object and read from input file
# If the input is the camera, pass 0 instead of the video file name
cap = cv2.VideoCapture('../data/video.mp4')

# allow the camera to warmup
time.sleep(0.1)

frame_count = 0

trackers = []

# Check if camera opened successfully
if (cap.isOpened()== False): 
  print("Error opening video stream or file")
 
# Read until video is completed
while(cap.isOpened()):
  # Capture frame-by-frame
    ret, frame = cap.read()
    if ret == True:
        frame = imutils.resize(frame, width=RES_W, height=RES_H)
        image = frame
        
        # loop over frames from the video stream
        # grab the raw NumPy array representing the image, then initialize the timestamp
        # and occupied/unoccupied text
        image_cpy = np.copy(image) # Create copy since cant modify orig

        if image_cpy is None:
            break

        # start the frames per second throughput estimator
        t1 = time.time()

        if frame_count == 0:
            _, bboxes = detectum.process_frame(image_cpy, THRESHOLD)

            # Loop through list (if empty this will be skipped) and overlay green bboxes
            for i in bboxes:
                print(i[0], i[1], i[2], i[3])
                cv2.rectangle(image_cpy, (i[0], i[1]), (i[2], i[3]), (0, 255, 0), 3)
                (startX, startY, endX, endY) = (i[0], i[1], i[2], i[3])

                # We need to initialize the tracker on the first frame
                # Create the correlation tracker - the object needs to be initialized
                # before it can be used
                tracker = dlib.correlation_tracker()

                # Start a track on face detected on first frame.
                rect = dlib.rectangle(i[0], i[1], i[2], i[3])
                tracker.start_track(image_cpy, rect)

                # add the tracker to our list of trackers so we can
                # utilize it during skip frames
                trackers.append(tracker)

        else:
            # Else we just attempt to track from the previous frame
            # track all the detected faces
            for tracker in trackers:
                tracker.update(image_cpy)
                pos = tracker.get_position()

                # unpack the position object
                startX = int(pos.left())
                startY = int(pos.top())
                endX = int(pos.right())
                endY = int(pos.bottom())

                # draw bounding box
                cv2.rectangle(image_cpy, (startX, startY), (endX, endY), (0, 255, 0), 3)

        frame_count += 1

        cv2.imshow('frame',image_cpy)
        key = cv2.waitKey(1) & 0xFF

        # if the `q` key was pressed, break from the loop
        if key == ord("q"):
            break

    else:
        break

print('Finished processing')
# When everything done, release the video capture object
cap.release()
out.release()

# Closes all the frames
cv2.destroyAllWindows()