コード例 #1
0
ファイル: main.py プロジェクト: andris955/diplomaterv
def main(algorithm: str, selected_mti: str, policy: str, train: bool = True,
         tb_log: bool = False, csv_log: bool = True, model_id: str = None):
    """
    :param algorithm: (str) 'A5C' or 'EA4C'
    :param selected_mti: (str) Name of the multi-task instance
     ['SpaceInvaders-v0', 'CrazyClimber-v0', 'Seaquest-v0', 'DemonAttack-v0', 'StarGunner-v0']'
    :param policy: (str) 'lstm' or 'ff' (feed forward)
    :param train: (bool) Whether to train or play
    :param tb_log: (bool) Whether to create tensorboard log or not. WARNING: IT LEAKS MEMORY IF IT IS TRUE
    :param csv_log: (bool) Whether or not to save results in csv files
    :param model_id: (str) or None Name of the model's directory which trying to load either for transfer learning or playing.
    """

    if isinstance(selected_mti, str):
        if selected_mti.lower() == 'mti1':
            selected_mti = config.MTI1
        elif selected_mti.lower() == 'mti2':
            selected_mti = config.MTI2
        elif selected_mti.lower() == 'mti3':
            selected_mti = config.MTI3
        elif selected_mti.lower() == 'mti4':
            selected_mti = config.MTI4
        elif selected_mti.lower() == 'mti5':
            selected_mti = config.MTI5
        elif selected_mti.lower() == 'mti6':
            selected_mti = config.MTI6
        elif selected_mti.lower() == 'mti7':
            selected_mti = config.MTI7
        elif selected_mti.lower() == 'mtic1':
            selected_mti = config.MTIC1
        elif selected_mti.lower() == 'mtic2':
            selected_mti = config.MTIC2
        elif selected_mti.lower() == 'mtic3':
            selected_mti = config.MTIC3
        else:
            selected_mti = config.MTIC1

    if train:
        n_cpus = cpu_count()
        dir_check()
        if tb_log:
            print("WARNING: TENSORBOARD LOGGING ACTIVE -> IT LEAKS MEMORY")
        repo = git.Repo(search_parent_directories=True)
        sha = repo.head.object.hexsha
        json_params = {
            'git_sha': sha,
        }
        mt = MultiTaskLearning(selected_mti, algorithm, policy, n_cpus, json_params,
                               csv_log, model_id, tensorboard_logging=tb_log)
        mt.train()
    else:
        if model_id == '':
            raise ValueError("If you want to play, you should provide a model")
        MultiTaskAgent.play(model_id, n_games=1, display=True)
コード例 #2
0
def dense_flow(fm):
    # initialize variables
    count = 0
    x = y = w = h = 0
    magnitude_histogram = []
    direction_histogram = []
    magnitude_histogram1 = []
    direction_histogram1 = []
    magnitude_histogram2 = []
    direction_histogram2 = []
    magnitude_histogram3 = []
    direction_histogram3 = []
    magnitude_histogram4 = []
    direction_histogram4 = []

    # start reading the video
    cap = cv2.VideoCapture(fm)

    # take the first frame and convert it to gray
    ret, frame = cap.read()
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

    # create the HSV color image
    hsvImg = np.zeros_like(frame)
    hsvImg[..., 1] = 255

    # play until the user decides to stop
    while True:
        # save the previous frame data
        previousGray = gray
        # get the next frame
        ret , frame = cap.read()
    
        if ret:
            # background-subtraction
            fgmask = fgbg.apply(frame)

            # median-blur
            seg_mask = cv2.medianBlur(fgmask, 5)

            # dilation
            seg_mask = cv2.dilate(seg_mask, kernel, iterations = 1)

            # for drawing bounding box over the entire body
            body = body_cascade.detectMultiScale(gray, 1.05, 3)
            if(len(body)!=0):
                for (x_t,y_t,w_t,h_t) in body:  
                    x, y, w, h = x_t, y_t, w_t, h_t
            
            # convert the frame to gray scale
            gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
            
            # exception-handling
            if((x, y, w, h) == (0 ,0, 0, 0)):
                continue

            # calculate the dense optical flow
            flow = cv2.calcOpticalFlowFarneback(previousGray, gray, None, 0.5, 3, 15, 3, 5, 1.2, 0)
        
            # obtain the flow magnitude and direction angle
            mag, ang = cv2.cartToPolar(flow[..., 0], flow[..., 1])
            mag = cv2.bitwise_and(mag, mag, mask = seg_mask)
            ang = cv2.bitwise_and(ang, ang, mask = seg_mask)

            # scaling
            ang=((ang*180)/(np.pi/2))%180
            
            # find the intersection points to draw the 2x2 grid
            k=1
            if(w%2==0):
                k=0
            c_x = x+(w//2)+k
            k=1
            if(h%2==0):
                k=0
            c_y = y+(h//2)+k

            flag1=flag2=flag3=flag4=0
            if(x-5>=0):
                x-=5
                flag1=1
                if(x+w+10<ang.shape[1]):
                    w+=10
                    flag2=1                    
            if(y-5>=0):
                y-=5
                flag3=1
                if(y+h+10<ang.shape[0]):
                    h+=10
                    flag4=1
            
            # extract the region-of-interests corresponding to the 2x2 grids
            roi_mag1 = mag[y:c_y, x:c_x]
            roi_mag2 = mag[y:c_y, c_x:x+w]
            roi_mag3 = mag[c_y:y+h, x:c_x]
            roi_mag4 = mag[c_y:y+h, c_x:x+w]
            roi_dir1 = ang[y:c_y, x:c_x]
            roi_dir2 = ang[y:c_y, c_x:x+w]
            roi_dir3 = ang[c_y:y+h, x:c_x]
            roi_dir4 = ang[c_y:y+h, c_x:x+w]

            magnitude = np.array(mag).flatten()
            direction = np.array(ang).flatten()
            magnitude1 = np.array(roi_mag1).flatten()
            direction1 = np.array(roi_dir1).flatten()
            magnitude2 = np.array(roi_mag2).flatten()
            direction2 = np.array(roi_dir2).flatten()
            magnitude3 = np.array(roi_mag3).flatten()
            direction3 = np.array(roi_dir3).flatten()
            magnitude4 = np.array(roi_mag4).flatten()
            direction4 = np.array(roi_dir4).flatten()

            # create magnitude and direction optical flow histogram per frame for each grid
            magnitude_histogram, direction_histogram = create_hist(magnitude, direction, magnitude_histogram, direction_histogram)
            magnitude_histogram1, direction_histogram1 = create_hist(magnitude1, direction1, magnitude_histogram1, direction_histogram1)
            magnitude_histogram2, direction_histogram2 = create_hist(magnitude2, direction2, magnitude_histogram2, direction_histogram2)
            magnitude_histogram3, direction_histogram3 = create_hist(magnitude3, direction3, magnitude_histogram3, direction_histogram3)
            magnitude_histogram4, direction_histogram4 = create_hist(magnitude4, direction4, magnitude_histogram4, direction_histogram4)
            
            #---------------------------------------------------------#
            # if you wish to see the optical flow frames uncomment the next 3 paragraphs             
            '''
            # update the color image
            hsvImg[..., 0] = 0.5 * ang * 180 / np.pi
            hsvImg[..., 2] = cv2.normalize(mag, None, 0, 255, cv2.NORM_MINMAX)
            rgbImg = cv2.cvtColor(hsvImg, cv2.COLOR_HSV2BGR)
            
            # drawing the bounding box
            cv2.rectangle(rgbImg, (x,y), (c_x,c_y), (255,0,0), 2)
            cv2.rectangle(rgbImg, (c_x,y), (x+w,c_y), (0,255,0), 2)
            cv2.rectangle(rgbImg, (x,c_y), (c_x,y+h), (0,0,255), 2)
            cv2.rectangle(rgbImg, (c_x,c_y), (x+w,y+h), (255,255,0), 2)
            
            # Display the resulting frame
            cv2.imshow('dense optical flow', np.hstack((frame, rgbImg)))
            '''
            #---------------------------------------------------------#

            # adjusting the bounding box over the POI to facilitate outward motion of the human body
            if(flag1==1):
                x+=5
                if(flag2==1):
                    w-=10
            if(flag3==1):
                y+=5
                if(flag4==1):
                    h-=10

            k = cv2.waitKey(30) & 0xff        
            if k == 27:
                break
        
        else:
            break


    # check the magnitude and direction histograms to have expected shapes
    magnitude_histogram = mag_check(magnitude_histogram)
    magnitude_histogram1 = mag_check(magnitude_histogram1)
    magnitude_histogram2 = mag_check(magnitude_histogram2)
    magnitude_histogram3 = mag_check(magnitude_histogram3)
    magnitude_histogram4 = mag_check(magnitude_histogram4)    
    direction_histogram = dir_check(direction_histogram)
    direction_histogram1 = dir_check(direction_histogram1)
    direction_histogram2 = dir_check(direction_histogram2)
    direction_histogram3 = dir_check(direction_histogram3)
    direction_histogram4 = dir_check(direction_histogram4) 

    # calculate the mean of the magnitude and direction histograms for each 2x2 grids
    mag_avg_hist = np.mean(magnitude_histogram, axis=0)
    dir_avg_hist = np.mean(direction_histogram, axis=0)
    mag_avg_hist1 = np.mean(magnitude_histogram1, axis=0)
    dir_avg_hist1 = np.mean(direction_histogram1, axis=0)
    mag_avg_hist2 = np.mean(magnitude_histogram2, axis=0)
    dir_avg_hist2 = np.mean(direction_histogram2, axis=0)
    mag_avg_hist3 = np.mean(magnitude_histogram3, axis=0)
    dir_avg_hist3 = np.mean(direction_histogram3, axis=0)
    mag_avg_hist4 = np.mean(magnitude_histogram4, axis=0)
    dir_avg_hist4 = np.mean(direction_histogram4, axis=0)

    # calculate the standard deviation of the magnitude and direction histograms for each 2x2 grids
    mag_std_hist = np.std(magnitude_histogram, axis=0)
    dir_std_hist = np.std(direction_histogram, axis=0)
    mag_std_hist1 = np.std(magnitude_histogram1, axis=0)
    dir_std_hist1 = np.std(direction_histogram1, axis=0)
    mag_std_hist2 = np.std(magnitude_histogram2, axis=0)
    dir_std_hist2 = np.std(direction_histogram2, axis=0)
    mag_std_hist3 = np.std(magnitude_histogram3, axis=0)
    dir_std_hist3 = np.std(direction_histogram3, axis=0)
    mag_std_hist4 = np.std(magnitude_histogram4, axis=0)
    dir_std_hist4 = np.std(direction_histogram4, axis=0)
 
    # concatenate all the histogram features to get the motion descriptor for 2x2 grids
    histogram = mag_avg_hist
    histogram = np.hstack((histogram, mag_std_hist))
    histogram = np.hstack((histogram, dir_avg_hist))
    histogram = np.hstack((histogram, dir_std_hist))
    histogram = np.hstack((histogram, mag_avg_hist1))
    histogram = np.hstack((histogram, mag_std_hist1))
    histogram = np.hstack((histogram, dir_avg_hist1))
    histogram = np.hstack((histogram, dir_std_hist1))
    histogram = np.hstack((histogram, mag_avg_hist2))
    histogram = np.hstack((histogram, mag_std_hist2))
    histogram = np.hstack((histogram, dir_avg_hist2))
    histogram = np.hstack((histogram, dir_std_hist2))
    histogram = np.hstack((histogram, mag_avg_hist3))
    histogram = np.hstack((histogram, mag_std_hist3))
    histogram = np.hstack((histogram, dir_avg_hist3))
    histogram = np.hstack((histogram, dir_std_hist3))
    histogram = np.hstack((histogram, mag_avg_hist4))
    histogram = np.hstack((histogram, mag_std_hist4))
    histogram = np.hstack((histogram, dir_avg_hist4))
    histogram = np.hstack((histogram, dir_std_hist4))

    cv2.destroyAllWindows()
    cap.release()
    return histogram
コード例 #3
0
def dense_flow(fm):
    # initialize variables
    count = 0
    x = y = w = h = 0
    magnitude_histogram = []
    direction_histogram = []
    magnitude_histogram1 = []
    direction_histogram1 = []
    magnitude_histogram2 = []
    direction_histogram2 = []
    magnitude_histogram3 = []
    direction_histogram3 = []
    magnitude_histogram4 = []
    direction_histogram4 = []
    magnitude_histogram5 = []
    direction_histogram5 = []
    magnitude_histogram6 = []
    direction_histogram6 = []
    magnitude_histogram7 = []
    direction_histogram7 = []
    magnitude_histogram8 = []
    direction_histogram8 = []
    magnitude_histogram9 = []
    direction_histogram9 = []
    magnitude_histogram10 = []
    direction_histogram10 = []
    magnitude_histogram11 = []
    direction_histogram11 = []
    magnitude_histogram12 = []
    direction_histogram12 = []
    magnitude_histogram13 = []
    direction_histogram13 = []
    magnitude_histogram14 = []
    direction_histogram14 = []
    magnitude_histogram15 = []
    direction_histogram15 = []
    magnitude_histogram16 = []
    direction_histogram16 = []

    # start reading the video
    cap = cv2.VideoCapture(fm)

    # take the first frame and convert it to gray
    ret, frame = cap.read()
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

    # create the HSV color image
    hsvImg = np.zeros_like(frame)
    hsvImg[..., 1] = 255

    # play until the user decides to stop
    frame_no = 0
    while True:
        # save the previous frame data
        previousGray = gray
        # get the next frame
        ret, frame = cap.read()

        if ret:
            # background-subtraction
            fgmask = fgbg.apply(frame)

            # median-blur
            seg_mask = cv2.medianBlur(fgmask, 5)

            # dilation
            seg_mask = cv2.dilate(seg_mask, kernel, iterations=1)

            #for drawing bounding box over the entire body
            body = body_cascade.detectMultiScale(gray, 1.05, 3)
            if (len(body) != 0):
                for (x_t, y_t, w_t, h_t) in body:
                    x, y, w, h = x_t, y_t, w_t, h_t

            # convert the frame to gray scale
            gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

            # exception-handling
            if ((x, y, w, h) == (0, 0, 0, 0)):
                continue

            # calculate the dense optical flow
            flow = cv2.calcOpticalFlowFarneback(previousGray, gray, None, 0.5,
                                                3, 15, 3, 5, 1.2, 0)

            # obtain the flow magnitude and direction angle
            mag, ang = cv2.cartToPolar(flow[..., 0], flow[..., 1])
            mag = cv2.bitwise_and(mag, mag, mask=seg_mask)
            ang = cv2.bitwise_and(ang, ang, mask=seg_mask)

            # scaling
            ang = ((ang * 180) / (np.pi / 2)) % 180

            # find the intersection points to draw the 4x4 grid
            k = 1
            if (w % 4 == 0):
                k = 0
            c_x1 = x + (w // 4) + k
            c_x2 = (x + 2 * (w // 4)) + k
            c_x3 = (x + 3 * (w // 4))

            k = 1
            if (h % 4 == 0):
                k = 0
            c_y1 = y + (h // 4) + k
            c_y2 = (y + 2 * (h // 4)) + k
            c_y3 = (y + 3 * (h // 4))

            flag1 = flag2 = flag3 = flag4 = 0
            if (x - 5 >= 0):
                x -= 5
                flag1 = 1
                if (x + w + 10 < ang.shape[1]):
                    w += 10
                    flag2 = 1
            if (y - 5 >= 0):
                y -= 5
                flag3 = 1
                if (y + h + 10 < ang.shape[0]):
                    h += 10
                    flag4 = 1

            # extract the region-of-interests corresponding to the 4x4 grids
            roi_mag1 = mag[y:c_y1, x:c_x1]
            roi_mag2 = mag[y:c_y1, c_x1:c_x2]
            roi_mag3 = mag[y:c_y1, c_x2:c_x3]
            roi_mag4 = mag[y:c_y1, c_x3:x + w]
            roi_mag5 = mag[c_y1:c_y2, x:c_x1]
            roi_mag6 = mag[c_y1:c_y2, c_x1:c_x2]
            roi_mag7 = mag[c_y1:c_y2, c_x2:c_x3]
            roi_mag8 = mag[c_y1:c_y2, c_x3:x + w]
            roi_mag9 = mag[c_y2:c_y3, x:c_x1]
            roi_mag10 = mag[c_y2:c_y3, c_x1:c_x2]
            roi_mag11 = mag[c_y2:c_y3, c_x2:c_x3]
            roi_mag12 = mag[c_y2:c_y3, c_x3:x + w]
            roi_mag13 = mag[c_y3:y + h, x:c_x1]
            roi_mag14 = mag[c_y3:y + h, c_x1:c_x2]
            roi_mag15 = mag[c_y3:y + h, c_x2:c_x3]
            roi_mag16 = mag[c_y3:y + h, c_x3:x + w]
            roi_dir1 = ang[y:c_y1, x:c_x1]
            roi_dir2 = ang[y:c_y1, c_x1:c_x2]
            roi_dir3 = ang[y:c_y1, c_x2:c_x3]
            roi_dir4 = ang[y:c_y1, c_x3:x + w]
            roi_dir5 = ang[c_y1:c_y2, x:c_x1]
            roi_dir6 = ang[c_y1:c_y2, c_x1:c_x2]
            roi_dir7 = ang[c_y1:c_y2, c_x2:c_x3]
            roi_dir8 = ang[c_y1:c_y2, c_x3:x + w]
            roi_dir9 = ang[c_y2:c_y3, x:c_x1]
            roi_dir10 = ang[c_y2:c_y3, c_x1:c_x2]
            roi_dir11 = ang[c_y2:c_y3, c_x2:c_x3]
            roi_dir12 = ang[c_y2:c_y3, c_x3:x + w]
            roi_dir13 = ang[c_y3:y + h, x:c_x1]
            roi_dir14 = ang[c_y3:y + h, c_x1:c_x2]
            roi_dir15 = ang[c_y3:y + h, c_x2:c_x3]
            roi_dir16 = ang[c_y3:y + h, c_x3:x + w]

            magnitude = np.array(mag).flatten()
            direction = np.array(ang).flatten()
            magnitude1 = np.array(roi_mag1).flatten()
            direction1 = np.array(roi_dir1).flatten()
            magnitude2 = np.array(roi_mag2).flatten()
            direction2 = np.array(roi_dir2).flatten()
            magnitude3 = np.array(roi_mag3).flatten()
            direction3 = np.array(roi_dir3).flatten()
            magnitude4 = np.array(roi_mag4).flatten()
            direction4 = np.array(roi_dir4).flatten()
            magnitude5 = np.array(roi_mag5).flatten()
            direction5 = np.array(roi_dir5).flatten()
            magnitude6 = np.array(roi_mag6).flatten()
            direction6 = np.array(roi_dir6).flatten()
            magnitude7 = np.array(roi_mag7).flatten()
            direction7 = np.array(roi_dir7).flatten()
            magnitude8 = np.array(roi_mag8).flatten()
            direction8 = np.array(roi_dir8).flatten()
            magnitude9 = np.array(roi_mag9).flatten()
            direction9 = np.array(roi_dir9).flatten()
            magnitude10 = np.array(roi_mag10).flatten()
            direction10 = np.array(roi_dir10).flatten()
            magnitude11 = np.array(roi_mag11).flatten()
            direction11 = np.array(roi_dir11).flatten()
            magnitude12 = np.array(roi_mag12).flatten()
            direction12 = np.array(roi_dir12).flatten()
            magnitude13 = np.array(roi_mag13).flatten()
            direction13 = np.array(roi_dir13).flatten()
            magnitude14 = np.array(roi_mag14).flatten()
            direction14 = np.array(roi_dir14).flatten()
            magnitude15 = np.array(roi_mag15).flatten()
            direction15 = np.array(roi_dir15).flatten()
            magnitude16 = np.array(roi_mag16).flatten()
            direction16 = np.array(roi_dir16).flatten()

            # create magnitude and direction optical flow histogram per frame for each grid
            magnitude_histogram, direction_histogram = create_hist_context(
                magnitude, direction, magnitude_histogram, direction_histogram)
            magnitude_histogram1, direction_histogram1 = create_hist_context(
                magnitude1, direction1, magnitude_histogram1,
                direction_histogram1)
            magnitude_histogram2, direction_histogram2 = create_hist_context(
                magnitude2, direction2, magnitude_histogram2,
                direction_histogram2)
            magnitude_histogram3, direction_histogram3 = create_hist_context(
                magnitude3, direction3, magnitude_histogram3,
                direction_histogram3)
            magnitude_histogram4, direction_histogram4 = create_hist_context(
                magnitude4, direction4, magnitude_histogram4,
                direction_histogram4)
            magnitude_histogram5, direction_histogram5 = create_hist_context(
                magnitude5, direction5, magnitude_histogram5,
                direction_histogram5)
            magnitude_histogram6, direction_histogram6 = create_hist_context(
                magnitude6, direction6, magnitude_histogram6,
                direction_histogram6)
            magnitude_histogram7, direction_histogram7 = create_hist_context(
                magnitude7, direction7, magnitude_histogram7,
                direction_histogram7)
            magnitude_histogram8, direction_histogram8 = create_hist_context(
                magnitude8, direction8, magnitude_histogram8,
                direction_histogram8)
            magnitude_histogram9, direction_histogram9 = create_hist_context(
                magnitude9, direction9, magnitude_histogram9,
                direction_histogram9)
            magnitude_histogram10, direction_histogram10 = create_hist_context(
                magnitude10, direction10, magnitude_histogram10,
                direction_histogram10)
            magnitude_histogram11, direction_histogram11 = create_hist_context(
                magnitude11, direction11, magnitude_histogram11,
                direction_histogram11)
            magnitude_histogram12, direction_histogram12 = create_hist_context(
                magnitude12, direction12, magnitude_histogram12,
                direction_histogram12)
            magnitude_histogram13, direction_histogram13 = create_hist_context(
                magnitude13, direction13, magnitude_histogram13,
                direction_histogram13)
            magnitude_histogram14, direction_histogram14 = create_hist_context(
                magnitude14, direction14, magnitude_histogram14,
                direction_histogram14)
            magnitude_histogram15, direction_histogram15 = create_hist_context(
                magnitude15, direction15, magnitude_histogram15,
                direction_histogram15)
            magnitude_histogram16, direction_histogram16 = create_hist_context(
                magnitude16, direction16, magnitude_histogram16,
                direction_histogram16)

            #---------------------------------------------------------#
            # if you wish to see the optical flow frames uncomment the next 3 paragraphs
            '''
            # update the color image
            hsvImg[..., 0] = 0.5 * ang * 180 / np.pi
            hsvImg[..., 2] = cv2.normalize(mag, None, 0, 255, cv2.NORM_MINMAX)
            rgbImg = cv2.cvtColor(hsvImg, cv2.COLOR_HSV2BGR)
            
            #drawing the bounding box
            cv2.rectangle(rgbImg, (x,y), (c_x1,c_y1), (255,0,0), 2)
            cv2.rectangle(rgbImg, (c_x1,y), (c_x2,c_y1), (0,255,0), 2)
            cv2.rectangle(rgbImg, (c_x2,y), (c_x3,c_y1), (0,0,255), 2)
            cv2.rectangle(rgbImg, (c_x3,y), (x+w,c_y1), (167,0,255), 2)

            cv2.rectangle(rgbImg, (x,c_y1), (c_x1,c_y2), (128,0,0), 2)
            cv2.rectangle(rgbImg, (c_x1, c_y1), (c_x2,c_y2), (0,128,0), 2)
            cv2.rectangle(rgbImg, (c_x2, c_y1), (c_x3,c_y2), (0,0,128), 2)
            cv2.rectangle(rgbImg, (c_x3, c_y1), (x+w,c_y2), (0,128,255), 2)

            cv2.rectangle(rgbImg, (x,c_y2), (c_x1,c_y3), (255,128,0), 2)
            cv2.rectangle(rgbImg, (c_x1, c_y2), (c_x2,c_y3), (0,255,128), 2)
            cv2.rectangle(rgbImg, (c_x2, c_y2), (c_x3,c_y3), (128,0,255), 2)
            cv2.rectangle(rgbImg, (c_x3, c_y2), (x+w,c_y3), (0,100,255), 2)

            cv2.rectangle(rgbImg, (x,c_y3), (c_x1,y+h), (255,0,128), 2)
            cv2.rectangle(rgbImg, (c_x1, c_y3), (c_x2,y+h), (128,255,128), 2)
            cv2.rectangle(rgbImg, (c_x2, c_y3), (c_x3,y+h), (128,128,255), 2)
            cv2.rectangle(rgbImg, (c_x3, c_y3), (x+w,y+h), (167,167,255), 2)
            
            #Display the resulting frame
            cv2.imshow('dense optical flow', np.hstack((frame, rgbImg)))
            '''
            #---------------------------------------------------------#

            # adjusting the bounding box over the POI to facilitate outward motion of the human body
            frame_no += 1
            if (flag1 == 1):
                x += 5
                if (flag2 == 1):
                    w -= 10
            if (flag3 == 1):
                y += 5
                if (flag4 == 1):
                    h -= 10

            k = cv2.waitKey(30) & 0xff
            if k == 27:
                break

        else:
            break

    # check the magnitude and direction histograms to have expected shapes
    magnitude_histogram = mag_check(magnitude_histogram)
    magnitude_histogram1 = mag_check(magnitude_histogram1)
    magnitude_histogram2 = mag_check(magnitude_histogram2)
    magnitude_histogram3 = mag_check(magnitude_histogram3)
    magnitude_histogram4 = mag_check(magnitude_histogram4)
    magnitude_histogram5 = mag_check(magnitude_histogram5)
    magnitude_histogram6 = mag_check(magnitude_histogram6)
    magnitude_histogram7 = mag_check(magnitude_histogram7)
    magnitude_histogram8 = mag_check(magnitude_histogram8)
    magnitude_histogram9 = mag_check(magnitude_histogram9)
    magnitude_histogram10 = mag_check(magnitude_histogram10)
    magnitude_histogram11 = mag_check(magnitude_histogram11)
    magnitude_histogram12 = mag_check(magnitude_histogram12)
    magnitude_histogram13 = mag_check(magnitude_histogram13)
    magnitude_histogram14 = mag_check(magnitude_histogram14)
    magnitude_histogram15 = mag_check(magnitude_histogram15)
    magnitude_histogram16 = mag_check(magnitude_histogram16)
    direction_histogram = dir_check(direction_histogram)
    direction_histogram1 = dir_check(direction_histogram1)
    direction_histogram2 = dir_check(direction_histogram2)
    direction_histogram3 = dir_check(direction_histogram3)
    direction_histogram4 = dir_check(direction_histogram4)
    direction_histogram5 = dir_check(direction_histogram5)
    direction_histogram6 = dir_check(direction_histogram6)
    direction_histogram7 = dir_check(direction_histogram7)
    direction_histogram8 = dir_check(direction_histogram8)
    direction_histogram9 = dir_check(direction_histogram9)
    direction_histogram10 = dir_check(direction_histogram10)
    direction_histogram11 = dir_check(direction_histogram11)
    direction_histogram12 = dir_check(direction_histogram12)
    direction_histogram13 = dir_check(direction_histogram13)
    direction_histogram14 = dir_check(direction_histogram14)
    direction_histogram15 = dir_check(direction_histogram15)
    direction_histogram16 = dir_check(direction_histogram16)

    # apply windowing to extract contextual information
    mag_hist = window(magnitude_histogram)
    dir_hist = window(direction_histogram)
    mag_hist1 = window(magnitude_histogram1)
    dir_hist1 = window(direction_histogram1)
    mag_hist2 = window(magnitude_histogram2)
    dir_hist2 = window(direction_histogram2)
    mag_hist3 = window(magnitude_histogram3)
    dir_hist3 = window(direction_histogram3)
    mag_hist4 = window(magnitude_histogram4)
    dir_hist4 = window(direction_histogram4)
    mag_hist5 = window(magnitude_histogram5)
    dir_hist5 = window(direction_histogram5)
    mag_hist6 = window(magnitude_histogram6)
    dir_hist6 = window(direction_histogram6)
    mag_hist7 = window(magnitude_histogram7)
    dir_hist7 = window(direction_histogram7)
    mag_hist8 = window(magnitude_histogram8)
    dir_hist8 = window(direction_histogram8)
    mag_hist9 = window(magnitude_histogram9)
    dir_hist9 = window(direction_histogram9)
    mag_hist10 = window(magnitude_histogram10)
    dir_hist10 = window(direction_histogram10)
    mag_hist11 = window(magnitude_histogram11)
    dir_hist11 = window(direction_histogram11)
    mag_hist12 = window(magnitude_histogram12)
    dir_hist12 = window(direction_histogram12)
    mag_hist13 = window(magnitude_histogram13)
    dir_hist13 = window(direction_histogram13)
    mag_hist14 = window(magnitude_histogram14)
    dir_hist14 = window(direction_histogram14)
    mag_hist15 = window(magnitude_histogram15)
    dir_hist15 = window(direction_histogram15)
    mag_hist16 = window(magnitude_histogram16)
    dir_hist16 = window(direction_histogram16)

    # calculate the mean of the magnitude and direction histograms for each 4x4 grids
    mag_avg_hist = np.mean(mag_hist, axis=0)
    dir_avg_hist = np.mean(dir_hist, axis=0)
    mag_avg_hist1 = np.mean(mag_hist1, axis=0)
    dir_avg_hist1 = np.mean(dir_hist1, axis=0)
    mag_avg_hist2 = np.mean(mag_hist2, axis=0)
    dir_avg_hist2 = np.mean(dir_hist2, axis=0)
    mag_avg_hist3 = np.mean(mag_hist3, axis=0)
    dir_avg_hist3 = np.mean(dir_hist3, axis=0)
    mag_avg_hist4 = np.mean(mag_hist4, axis=0)
    dir_avg_hist4 = np.mean(dir_hist4, axis=0)
    mag_avg_hist5 = np.mean(mag_hist5, axis=0)
    dir_avg_hist5 = np.mean(dir_hist5, axis=0)
    mag_avg_hist6 = np.mean(mag_hist6, axis=0)
    dir_avg_hist6 = np.mean(dir_hist6, axis=0)
    mag_avg_hist7 = np.mean(mag_hist7, axis=0)
    dir_avg_hist7 = np.mean(dir_hist7, axis=0)
    mag_avg_hist8 = np.mean(mag_hist8, axis=0)
    dir_avg_hist8 = np.mean(dir_hist8, axis=0)
    mag_avg_hist9 = np.mean(mag_hist9, axis=0)
    dir_avg_hist9 = np.mean(dir_hist9, axis=0)
    mag_avg_hist10 = np.mean(mag_hist10, axis=0)
    dir_avg_hist10 = np.mean(dir_hist10, axis=0)
    mag_avg_hist11 = np.mean(mag_hist11, axis=0)
    dir_avg_hist11 = np.mean(dir_hist11, axis=0)
    mag_avg_hist12 = np.mean(mag_hist12, axis=0)
    dir_avg_hist12 = np.mean(dir_hist12, axis=0)
    mag_avg_hist13 = np.mean(mag_hist13, axis=0)
    dir_avg_hist13 = np.mean(dir_hist13, axis=0)
    mag_avg_hist14 = np.mean(mag_hist14, axis=0)
    dir_avg_hist14 = np.mean(dir_hist14, axis=0)
    mag_avg_hist15 = np.mean(mag_hist15, axis=0)
    dir_avg_hist15 = np.mean(dir_hist15, axis=0)
    mag_avg_hist16 = np.mean(mag_hist16, axis=0)
    dir_avg_hist16 = np.mean(dir_hist16, axis=0)

    # calculate the standard deviation of the magnitude and direction histograms for each 4x4 grids
    mag_std_hist = np.std(mag_hist, axis=0)
    dir_std_hist = np.std(dir_hist, axis=0)
    mag_std_hist1 = np.std(mag_hist1, axis=0)
    dir_std_hist1 = np.std(dir_hist1, axis=0)
    mag_std_hist2 = np.std(mag_hist2, axis=0)
    dir_std_hist2 = np.std(dir_hist2, axis=0)
    mag_std_hist3 = np.std(mag_hist3, axis=0)
    dir_std_hist3 = np.std(dir_hist3, axis=0)
    mag_std_hist4 = np.std(mag_hist4, axis=0)
    dir_std_hist4 = np.std(dir_hist4, axis=0)
    mag_std_hist5 = np.std(mag_hist5, axis=0)
    dir_std_hist5 = np.std(dir_hist5, axis=0)
    mag_std_hist6 = np.std(mag_hist6, axis=0)
    dir_std_hist6 = np.std(dir_hist6, axis=0)
    mag_std_hist7 = np.std(mag_hist7, axis=0)
    dir_std_hist7 = np.std(dir_hist7, axis=0)
    mag_std_hist8 = np.std(mag_hist8, axis=0)
    dir_std_hist8 = np.std(dir_hist8, axis=0)
    mag_std_hist9 = np.std(mag_hist9, axis=0)
    dir_std_hist9 = np.std(dir_hist9, axis=0)
    mag_std_hist10 = np.std(mag_hist10, axis=0)
    dir_std_hist10 = np.std(dir_hist10, axis=0)
    mag_std_hist11 = np.std(mag_hist11, axis=0)
    dir_std_hist11 = np.std(dir_hist11, axis=0)
    mag_std_hist12 = np.std(mag_hist12, axis=0)
    dir_std_hist12 = np.std(dir_hist12, axis=0)
    mag_std_hist13 = np.std(mag_hist13, axis=0)
    dir_std_hist13 = np.std(dir_hist13, axis=0)
    mag_std_hist14 = np.std(mag_hist14, axis=0)
    dir_std_hist14 = np.std(dir_hist14, axis=0)
    mag_std_hist15 = np.std(mag_hist15, axis=0)
    dir_std_hist15 = np.std(dir_hist15, axis=0)
    mag_std_hist16 = np.std(mag_hist16, axis=0)
    dir_std_hist16 = np.std(dir_hist16, axis=0)

    # concatenate all the histogram features to get the contextual descriptor for 4x4 grids
    histogram = mag_avg_hist
    histogram = np.hstack((histogram, mag_std_hist))
    histogram = np.hstack((histogram, dir_avg_hist))
    histogram = np.hstack((histogram, dir_std_hist))
    histogram = np.hstack((histogram, mag_avg_hist1))
    histogram = np.hstack((histogram, mag_std_hist1))
    histogram = np.hstack((histogram, dir_avg_hist1))
    histogram = np.hstack((histogram, dir_std_hist1))
    histogram = np.hstack((histogram, mag_avg_hist2))
    histogram = np.hstack((histogram, mag_std_hist2))
    histogram = np.hstack((histogram, dir_avg_hist2))
    histogram = np.hstack((histogram, dir_std_hist2))
    histogram = np.hstack((histogram, mag_avg_hist3))
    histogram = np.hstack((histogram, mag_std_hist3))
    histogram = np.hstack((histogram, dir_avg_hist3))
    histogram = np.hstack((histogram, dir_std_hist3))
    histogram = np.hstack((histogram, mag_avg_hist4))
    histogram = np.hstack((histogram, mag_std_hist4))
    histogram = np.hstack((histogram, dir_avg_hist4))
    histogram = np.hstack((histogram, dir_std_hist4))
    histogram = np.hstack((histogram, mag_avg_hist5))
    histogram = np.hstack((histogram, mag_std_hist5))
    histogram = np.hstack((histogram, dir_avg_hist5))
    histogram = np.hstack((histogram, dir_std_hist5))
    histogram = np.hstack((histogram, mag_avg_hist6))
    histogram = np.hstack((histogram, mag_std_hist6))
    histogram = np.hstack((histogram, dir_avg_hist6))
    histogram = np.hstack((histogram, dir_std_hist6))
    histogram = np.hstack((histogram, mag_avg_hist7))
    histogram = np.hstack((histogram, mag_std_hist7))
    histogram = np.hstack((histogram, dir_avg_hist7))
    histogram = np.hstack((histogram, dir_std_hist7))
    histogram = np.hstack((histogram, mag_avg_hist8))
    histogram = np.hstack((histogram, mag_std_hist8))
    histogram = np.hstack((histogram, dir_avg_hist8))
    histogram = np.hstack((histogram, dir_std_hist8))
    histogram = np.hstack((histogram, mag_avg_hist9))
    histogram = np.hstack((histogram, mag_std_hist9))
    histogram = np.hstack((histogram, dir_avg_hist9))
    histogram = np.hstack((histogram, dir_std_hist9))
    histogram = np.hstack((histogram, mag_avg_hist10))
    histogram = np.hstack((histogram, mag_std_hist10))
    histogram = np.hstack((histogram, dir_avg_hist10))
    histogram = np.hstack((histogram, dir_std_hist10))
    histogram = np.hstack((histogram, mag_avg_hist11))
    histogram = np.hstack((histogram, mag_std_hist11))
    histogram = np.hstack((histogram, dir_avg_hist11))
    histogram = np.hstack((histogram, dir_std_hist11))
    histogram = np.hstack((histogram, mag_avg_hist12))
    histogram = np.hstack((histogram, mag_std_hist12))
    histogram = np.hstack((histogram, dir_avg_hist12))
    histogram = np.hstack((histogram, dir_std_hist12))
    histogram = np.hstack((histogram, mag_avg_hist13))
    histogram = np.hstack((histogram, mag_std_hist13))
    histogram = np.hstack((histogram, dir_avg_hist13))
    histogram = np.hstack((histogram, dir_std_hist13))
    histogram = np.hstack((histogram, mag_avg_hist14))
    histogram = np.hstack((histogram, mag_std_hist14))
    histogram = np.hstack((histogram, dir_avg_hist14))
    histogram = np.hstack((histogram, dir_std_hist14))
    histogram = np.hstack((histogram, mag_avg_hist15))
    histogram = np.hstack((histogram, mag_std_hist15))
    histogram = np.hstack((histogram, dir_avg_hist15))
    histogram = np.hstack((histogram, dir_std_hist15))
    histogram = np.hstack((histogram, mag_avg_hist16))
    histogram = np.hstack((histogram, mag_std_hist16))
    histogram = np.hstack((histogram, dir_avg_hist16))
    histogram = np.hstack((histogram, dir_std_hist16))

    cv2.destroyAllWindows()
    cap.release()
    return histogram, frame_no
コード例 #4
0
# 1. Contstruct the path to the text file in the data directory using the `pathlib` module [2P]

cars_dir = Path(data_dir, "cars.txt")

# 2. Read the text file [2P]

cars = open(str(cars_dir), "r").read()

# 3. Count the occurences of each item in the text file [2P]

clist = list(cars.split("\n"))
occur = ut.countelem(clist)

# 4. Using `pathlib` check if a directory with name `solution` exists and if not create it [2P]

ut.dir_check(output_dir)

# 5. Write the counts to the file `counts.csv` in the `solution` directory in the format (first line is the header): [2P]
#    item, count
#    item_name_1, item_count_1
#    item_name_2, item_count_2
#    ...

counts = output_dir / "counts.csv"

with open(counts, "w") as file:
    writer = csv.DictWriter(file, fieldnames=["item", "count"])
    writer.writeheader()
    for key in occur.keys():
        file.write("%s,%s\n" % (key, occur[key]))