Beispiel #1
0
def pdt_init_constants():
    global ASPECT_RATIO, PADDING, SCALE, WIN_STRIDE, BORDER_AROUND_BLOB, \
        PERSON_DETECTION_PARALLEL_MODE

    # Configuration parameters
    ASPECT_RATIO = config.getfloat('ASPECT_RATIO')
    PADDING = (config.getint('PADDING_0'), config.getint('PADDING_1'))
    SCALE = config.getfloat('SCALE')
    WIN_STRIDE = (config.getint('WINSTRIDE_0'), config.getint('WINSTRIDE_1'))
    BORDER_AROUND_BLOB = (config.getfloat('BORDER_AROUND_BLOB_0'),
                          config.getfloat('BORDER_AROUND_BLOB_1'))
    PERSON_DETECTION_PARALLEL_MODE = \
        config.getboolean("PERSON_DETECTION_PARALLEL_MODE")
    def __init__(self, shape):
        """  START SETTING CONSTANTS  """

        global USE_CONFIDENCE_LEVELS, CONFIDENCE_LEVELS, \
            USE_SQUARE_REGION_FOR_VERIFY, SQUARE_REGION_RADIUS

        USE_CONFIDENCE_LEVELS = config.getboolean("USE_CONFIDENCE_LEVELS")
        CONFIDENCE_LEVELS = (config.getfloat("CONFIDENCE_LEVEL_0"),
                             config.getfloat("CONFIDENCE_LEVEL_1"))
        USE_SQUARE_REGION_FOR_VERIFY = \
            config.getboolean("USE_SQUARE_REGION_FOR_VERIFY")
        SQUARE_REGION_RADIUS = config.getint("SQUARE_REGION_RADIUS")
        """  FINISH SETTING CONSTANTS  """

        self.confidenceMatrix = np.zeros(shape=(int(shape[0] / 5),
                                                int(shape[1] / 5)),
                                         dtype=np.float64)
        self.xedges =\
            np.linspace(0, int(self.confidenceMatrix.shape[0] * 5),
                        int(self.confidenceMatrix.shape[0]) + 1,
                        endpoint=True)
        self.yedges =\
            np.linspace(0, int(self.confidenceMatrix.shape[1] * 5),
                        int(self.confidenceMatrix.shape[1]) + 1,
                        endpoint=True)
        self.normalizedConfidenceMatrix = np.zeros_like(self.confidenceMatrix,
                                                        dtype=np.float64)
        self.onePersonConfidenceMatrix = np.zeros_like(self.confidenceMatrix,
                                                       dtype=np.float64)
        self.updateMaximums = \
            np.zeros_like(self.confidenceMatrix, dtype=np.int)
Beispiel #3
0
    def __init__(self):
        self.gaussian_size = (config.getint('GAUSSIANBLUR_SIZE_X'),
                              config.getint('GAUSSIANBLUR_SIZE_Y'))
        self.erode_size = np.ones(
            (config.getint('ERODE_SIZE_X'), config.getint('ERODE_SIZE_Y')),
            np.uint8)
        self.erode_times = config.getint('ERODE_TIMES')
        self.dilate_size = np.ones(
            (config.getint('DILATE_SIZE_X'), config.getint('DILATE_SIZE_Y')),
            np.uint8)
        self.dilate_times = config.getint('DILATE_TIMES')
        self.history = config.getint('HISTORY')
        self.detect_shadows = config.getboolean('DETECT_SHADOWS')
        self.learning_rate = config.getfloat('MOG2_LEARNING_RATE')

        self.subtractor = cv2.createBackgroundSubtractorMOG2(
            history=self.history, detectShadows=self.detect_shadows)
def pd_init_constants():
    global BORDER_AROUND_BLOB, USE_HISTOGRAMS_FOR_PERSON_DETECTION, \
        FRAMES_COUNT_FOR_TRAINING_HISTOGRAMS, CONFIDENCE_MATRIX_UPDATE_TIME, \
        PERSON_DETECTION_PARALLEL_MODE, PROCESSES_POOL

    # LOAD CONFIG. PARAMETERS
    BORDER_AROUND_BLOB = (config.getfloat("BORDER_AROUND_BLOB_0"),
                          config.getfloat("BORDER_AROUND_BLOB_1"))
    USE_HISTOGRAMS_FOR_PERSON_DETECTION = \
        config.getboolean("USE_HISTOGRAMS_FOR_PERSON_DETECTION")
    FRAMES_COUNT_FOR_TRAINING_HISTOGRAMS = \
        config.getint("FRAMES_COUNT_FOR_TRAINING_HISTOGRAMS")
    CONFIDENCE_MATRIX_UPDATE_TIME = \
        config.getint("CONFIDENCE_MATRIX_UPDATE_TIME")
    PERSON_DETECTION_PARALLEL_MODE = \
        config.getboolean("PERSON_DETECTION_PARALLEL_MODE")

    if PERSON_DETECTION_PARALLEL_MODE:
        from multiprocessing.pool import Pool
        PROCESSES_POOL = Pool()

    pdt_init_constants()
def get_status_info_comm():
    STATUS_INFO_EXCHANGE_HOSTADDRESS = \
        config.get('STATUS_INFO_EXCHANGE_HOSTADDRESS')
    STATUS_INFO_EXCHANGE_NAME = \
        config.get('STATUS_INFO_EXCHANGE_NAME')
    STATUS_INFO_EXPIRATION_TIME = \
        config.getint('STATUS_INFO_EXPIRATION_TIME')
    comm_info = Communicator(host_address=STATUS_INFO_EXCHANGE_HOSTADDRESS,
                             exchange=STATUS_INFO_EXCHANGE_NAME,
                             exchange_type='topic',
                             expiration_time=STATUS_INFO_EXPIRATION_TIME)

    return comm_info
    def __init__(self):

        # Configuration parameters

        self.expand_blobs = (config.getboolean('EXPAND_BLOBS'),
                             config.getfloat('EXPAND_BLOBS_RATIO'))

        self.detect_blobs_by_bounding_boxes = \
            config.getboolean('DETECT_BLOBS_BY_BOUNDING_BOXES')

        # If SimpleBlobDetector will be used, load the parameters
        if not self.detect_blobs_by_bounding_boxes:

            self.threshold = [
                config.getint('MIN_THRESHOLD'),
                config.getint('MAX_THRESHOLD'),
                config.getint('THRESHOLD_STEP')
            ]
            self.min_dist_between_blobs = \
                config.getint('MIN_DIST_BETWEEN_BLOBS')
            self.filter_by_color = [
                config.getboolean('FILTER_BY_COLOR'),
                config.getint('BLOB_COLOR')
            ]
            self.filter_by_area = [
                config.getboolean('FILTER_BY_AREA'),
                config.getint('MIN_AREA'),
                config.getint('MAX_AREA')
            ]
            self.filter_by_circularity = \
                [config.getboolean('FILTER_BY_CIRCULARITY'),
                 config.getfloat('MIN_CIRCULARITY'),
                 config.getfloat('MAX_CIRCULARITY')]
            self.filter_by_convexity = [
                config.getboolean('FILTER_BY_CONVEXITY'),
                config.getfloat('MIN_CONVEXITY'),
                config.getfloat('MAX_CONVEXITY')
            ]
            self.filter_by_inertia = [
                config.getboolean('FILTER_BY_INERTIA'),
                config.getfloat('MIN_INERTIA'),
                config.getfloat('MAX_INERTIA')
            ]

            # Setup SimpleBlobDetector parameters
            params = cv2.SimpleBlobDetector_Params()

            # Change thresholds
            params.minThreshold = self.threshold[0]
            params.maxThreshold = self.threshold[1]
            params.thresholdStep = self.threshold[2]

            # Minimum distance between blobs
            params.minDistBetweenBlobs = self.min_dist_between_blobs

            # Filter by Color
            params.filterByColor = self.filter_by_color[0]
            params.blobColor = self.filter_by_color[1]

            # Filter by Area.
            params.filterByArea = self.filter_by_area[0]
            params.minArea = self.filter_by_area[1]
            params.maxArea = self.filter_by_area[2]

            # Filter by Circularity
            params.filterByCircularity = self.filter_by_circularity[0]
            params.minCircularity = self.filter_by_circularity[1]
            params.maxCircularity = self.filter_by_circularity[2]

            # Filter by Convexity
            params.filterByConvexity = self.filter_by_convexity[0]
            params.minConvexity = self.filter_by_convexity[1]
            params.maxConvexity = self.filter_by_convexity[2]

            # Filter by Inertia
            params.filterByInertia = self.filter_by_inertia[0]
            params.minInertiaRatio = self.filter_by_inertia[1]
            params.maxInertiaRatio = self.filter_by_inertia[2]

            self.detector = cv2.SimpleBlobDetector_create(params)
Beispiel #7
0
    def __init__(self):

        # Configuration parameters
        self.gaussian_size = (config.getint('GAUSSIANBLUR_SIZE_X'),
                              config.getint('GAUSSIANBLUR_SIZE_Y'))
        self.erode_size = np.ones(
            (config.getint('ERODE_SIZE_X'), config.getint('ERODE_SIZE_Y')),
            np.uint8)
        self.erode_times = config.getint('ERODE_TIMES')
        self.dilate_size = np.ones(
            (config.getint('DILATE_SIZE_X'), config.getint('DILATE_SIZE_Y')),
            np.uint8)
        self.dilate_times = config.getint('DILATE_TIMES')

        self.history = config.getint('HISTORY')
        self.dist_2_threshold = config.getint('DIST_2_THRESHOLD')
        self.n_samples = config.getint('N_SAMPLES')
        self.knn_samples = config.getint('KNN_SAMPLES')
        self.detect_shadows = config.getboolean('DETECT_SHADOWS')
        self.shadow_threshold = config.getfloat('SHADOW_THRESHOLD')

        self.subtractor = cv2.createBackgroundSubtractorKNN()

        # Sets the number of last frames that affect the background model.
        self.subtractor.setHistory(self.history)

        # Sets the threshold on the squared distance between the pixel and
        # the sample. The threshold on the squared distance between the
        # pixel and the sample to decide \
        # whether a pixel is close to a data sample.
        self.subtractor.setDist2Threshold(self.dist_2_threshold)

        # Sets the shadow detection flag. \
        # If true, the algorithm detects shadows and marks them.
        self.subtractor.setDetectShadows(self.detect_shadows)

        # Sets the number of neighbours, the k in kNN. \
        # K is the number of samples that need to be within dist2Threshold
        # in order \
        # to decide that that pixel is matching the kNN background model.
        # Sets the k in the kNN. How many nearest neighbors need to match.
        self.subtractor.setkNNSamples(self.knn_samples)

        # Sets the shadow threshold. A shadow is detected if pixel is a
        # darker version of the background. The shadow threshold
        # (Tau in the paper) is a threshold defining how much darker the
        # shadow can be. Tau= 0.5 means that if a pixel is more than twice \
        # darker then it is not shadow. See Prati, Mikic,
        # Trivedi and Cucchiarra, \
        # Detecting Moving Shadows...*, IEEE PAMI,2003.
        self.subtractor.setShadowThreshold(self.shadow_threshold)
        self.subtractor.setShadowValue(0)

        # Sets the number of data samples in the background model. \
        # The model needs to be reinitialized to reserve memory.
        self.subtractor.setNSamples(self.n_samples)
def track_source(identifier=None, source=None, trackermaster_conf=None,
                 patternmaster_conf=None):
    """
    :param identifier:
    :param source:
    :param trackermaster_conf:
    :param patternmaster_conf:
    :return:
    """

    """  START SETTING CONSTANTS  """
    trackermaster_conf=None
    patternmaster_conf=None

    global USE_HISTOGRAMS_FOR_PERSON_DETECTION, SHOW_PREDICTION_DOTS, \
        SHOW_COMPARISONS_BY_COLOR, SHOW_VIDEO_OUTPUT, LIMIT_FPS, \
        DEFAULT_FPS_LIMIT, CREATE_MODEL, USE_MODEL, SAVE_POSITIONS_TO_FILE, \
        VERBOSE

    USE_HISTOGRAMS_FOR_PERSON_DETECTION = \
        config.getboolean("USE_HISTOGRAMS_FOR_PERSON_DETECTION")
    SHOW_PREDICTION_DOTS = config.getboolean("SHOW_PREDICTION_DOTS")
    SHOW_COMPARISONS_BY_COLOR = config.getboolean("SHOW_COMPARISONS_BY_COLOR")
    SHOW_VIDEO_OUTPUT = config.getboolean("SHOW_VIDEO_OUTPUT")
    LIMIT_FPS = config.getboolean("LIMIT_FPS")
    DEFAULT_FPS_LIMIT = config.getfloat("DEFAULT_FPS_LIMIT")
    if CREATE_MODEL is None:
        CREATE_MODEL = config.getboolean("CREATE_MODEL")
    if USE_MODEL is None:
        USE_MODEL = config.getboolean("USE_MODEL")
    SAVE_POSITIONS_TO_FILE = config.getboolean("SAVE_POSITIONS_TO_FILE")
    VERBOSE = config.getboolean('VERBOSE')
    USE_BSUBTRACTOR_KNN = config.getboolean("USE_BSUBTRACTOR_KNN")

    """  FINISH SETTING CONSTANTS  """

    if not identifier:
        identifier = sha1(str(dt.utcnow()).encode('utf-8')).hexdigest()
    if trackermaster_conf:
        set_custome_config(trackermaster_conf)

    # Instance of VideoCapture to capture webcam(0) images
    # WebCam
    # cap = cv2.VideoCapture(0)
    # popen("v4l2-ctl -d /dev/video1 --set-ctrl "
    #       "white_balance_temperature_auto=0,"
    #       "white_balance_temperature=inactive,exposure_absolute=inactive,"
    #       "focus_absolute=inactive,focus_auto=0,exposure_auto_priority=0")

    # Communication with Launcher and others
    comm_info = get_status_info_comm()

    # Communication with PatternMaster
    communicator = \
        Communicator(exchange=config.get('TRACK_INFO_EXCHANGE_NAME'),
                     host_address=config.get(
                         'TRACK_INFO_EXCHANGE_HOSTADDRESS'),
                     expiration_time=config.getint(
                         'TRACK_INFO_EXPIRATION_TIME'),
                     exchange_type='direct')
    exit_cause = 'FINISHED'

    global cap
    global has_more_images
    global raw_image
    global processed
    global SEC_PER_FRAME

    if source:

        cap = cv2.VideoCapture(source)
        comm_info.send_message(
            json.dumps(dict(
                info_id="OPEN", id=identifier,
                content="Opening source: %s." % source)),
            routing_key='info')
    else:
        # Videos de muestra
        videos_path = os.path.dirname(
            os.path.abspath(inspect.getfile(inspect.currentframe())))
        # source = videos_path + '/../Videos/Video_003.avi'
        source = videos_path + '/../../videos_demo/DSC_3133_luminosa.mkv'
        cap = cv2.VideoCapture(source)

    has_at_least_one_frame, raw_image = cap.read()

    if not has_at_least_one_frame:
        comm_info.send_message(json.dumps(dict(
            info_id="EXIT WITH ERROR", id=identifier,
            content="<p>ERROR: Trying to open source but couldn't.</p>")),
            routing_key='info')
        print('EXIT %s with error: Source %s could not be loaded.' %
              (identifier, source))

        exit()

    # Original FPS
    try:
        FPS = float(int(cap.get(cv2.CAP_PROP_FPS)))
        if FPS == 0.:
            FPS = DEFAULT_FPS_LIMIT
    except ValueError:
        FPS = DEFAULT_FPS_LIMIT

    reader = Thread(target=read_raw_input, daemon=True)
    reader.start()

    print("Working at", FPS, "FPS")
    SEC_PER_FRAME = 1. / FPS
    FPS_OVER_2 = (FPS / 2)

    # Getting width and height of captured images
    w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
    h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
    print("Real resolution: Width", w, "Height", h)
    resolution_multiplier = find_resolution_multiplier(w, h)
    work_w = int(w / resolution_multiplier)
    work_h = int(h / resolution_multiplier)
    print("Work resolution: Width", work_w, "Height", work_h)

    send_patternrecognition_config(communicator, identifier,
                                   patternmaster_conf, resolution_multiplier)

    font = cv2.FONT_HERSHEY_SIMPLEX

    background_subtractor = BackgroundSubtractorKNN() if USE_BSUBTRACTOR_KNN \
        else BackgroundSubtractorMOG2()

    blobs_detector = BlobDetector()
    # person_detector = Histogram2D()
    person_detection.pd_init_constants()
    tracker = Tracker(FPS, resolution_multiplier)

    loop_time = time.time()

    global number_frame
    _fps = "%.2f" % FPS
    previous_fps = FPS

    read_time = 0
    max_read_time = 0
    bg_sub_time = 0
    max_bg_sub_time = 0
    blob_det_time = 0
    max_blob_det_time = 0
    person_detection_time = 0
    max_person_detection_time = 0
    t_time = 0
    max_t_time = 0
    pattern_recogn_time = 0
    max_pattern_recogn_time = 0
    show_info_time = 0
    max_show_info_time = 0
    display_time = 0
    max_display_time = 0
    wait_key_time = 0
    max_wait_key_time = 0
    total_time = 0
    max_total_time = 0

    persons_in_scene = "Frame number (one-based), Current persons detected, " \
                       "Current tracklets, " \
                       "Current tracklets/persons interpol. num\n\n"

    model_load = True, ""
    if USE_HISTOGRAMS_FOR_PERSON_DETECTION:
        person_detection.set_histogram_size(shape=(work_w, work_h))
        person_detection.set_create_model(CREATE_MODEL)
        model_load = person_detection.set_use_model(USE_MODEL)

    fps = 0
    comparisons_by_color_image = []
    positions_to_file = ''
    interpol_cant_persons_prev = 0
    trayectos = []
    tracklets = {}
    last_number_frame = number_frame
    p_matrix_history = ''

    if model_load[0]:
        # Start the main loop
        while has_more_images:

            t_total = time.time()

            # FPS calculation
            if number_frame > 10 and number_frame != last_number_frame:
                delay = (time.time() - loop_time)
                loop_time = time.time()
                # if LIMIT_FPS:
                #     if delay < SEC_PER_FRAME:
                #         time_aux = time.time()
                #         time.sleep(max(SEC_PER_FRAME - delay, 0))
                #         delay += time.time() - time_aux

                fps = (1. / delay) * 0.25 + previous_fps * 0.75
                previous_fps = fps
                _fps = "%.2f" % fps

            else:
                if LIMIT_FPS:
                    while has_more_images and \
                            number_frame == last_number_frame:
                        time.sleep(0.01)  # Sleep for avoid Busy waiting
                    if not has_more_images:
                        break
                    loop_time = time.time()

            t0 = time.time()

            aux_time = time.time() - t0
            if number_frame > 200:
                read_time += aux_time
                max_read_time = max(aux_time, max_read_time)

            if has_more_images:
                # ########################################################## #
                # ##               BLACK BOXES PROCESSES                  ## #
                # ########################################################## #

                # ########################## ##
                # ## BACKGROUND SUBTRACTOR # ##
                # ########################## ##

                t0 = time.time()

                # resize to a manageable work resolution
                if LIMIT_FPS:
                    reader_lock.acquire()
                else:
                    reader_condition.acquire()
                    if number_frame == last_number_frame and has_more_images:
                        reader_condition.wait(2)

                if not has_more_images:
                    if LIMIT_FPS:
                        reader_lock.release()
                    else:
                        reader_condition.notify()
                        reader_condition.release()
                    break
                else:
                    last_number_frame = number_frame
                    raw_frame_copy = raw_image.copy()
                if LIMIT_FPS:
                    reader_lock.release()
                else:
                    processed = True
                    reader_condition.notify()
                    reader_condition.release()

                frame_resized = cv2.resize(raw_frame_copy, (work_w, work_h))
                frame_resized_copy = frame_resized.copy()

                bg_sub = background_subtractor.apply(frame_resized)
                bg_subtraction = cv2.cvtColor(bg_sub, cv2.COLOR_GRAY2BGR)
                to_show = bg_subtraction.copy()

                bg_subtraction_resized =\
                    cv2.resize(bg_subtraction, (work_w, work_h))

                aux_time = time.time() - t0
                if number_frame > 200:
                    bg_sub_time += aux_time
                    max_bg_sub_time = max(aux_time, max_bg_sub_time)

                # ################### ##
                # ## BLOBS DETECTOR # ##
                # ################### ##

                t0 = time.time()

                bounding_boxes = blobs_detector.apply(bg_sub)

                aux_time = time.time() - t0
                if number_frame > 200:
                    blob_det_time += aux_time
                    max_blob_det_time = max(aux_time, max_blob_det_time)

                t0 = time.time()

                cant_personas = 0

                if len(bounding_boxes):
                    rectangles = x1y1x2y2_to_x1y1wh(bounding_boxes)
                    del bounding_boxes

                    for (x, y, w, h) in rectangles:
                        # Draw in blue candidate blobs
                        cv2.rectangle(frame_resized_copy, (x, y),
                                      (x + w, y + h), (255, 0, 0), 1)

                    if len(rectangles) > 100:
                        # Skip the cycle when it's full of small blobs
                        continue

                    # ##################### ##
                    # ## PERSONS DETECTOR # ##
                    # ##################### ##
                    persons = person_detection.apply(
                        rectangles, resolution_multiplier, raw_frame_copy,
                        frame_resized_copy, number_frame, fps)
                    cant_personas = len(persons)

                    for p in persons:
                        # Red and Yellow dots
                        (x_a, y_a), (x_b, y_b) = p['box']
                        color = 0 if p['score'] == 1 else 255
                        cv2.circle(img=frame_resized_copy,
                                   center=(int((x_a + x_b) / 2),
                                           int((y_a + y_b) / 2)), radius=0,
                                   color=(0, color, 255), thickness=3)

                    aux_time = time.time() - t0
                    if number_frame > 200:
                        person_detection_time += aux_time
                        max_person_detection_time = \
                            max(aux_time, max_person_detection_time)

                    t0 = time.time()

                    # ############ ##
                    # ## TRACKER # ##
                    # ############ ##
                    rectangles_in_frame = []
                    trayectos_, info_to_send, tracklets, \
                        comparisons_by_color_image_aux, \
                        positions_in_frame,\
                        rectangles_in_frame,\
                        frame_p_matrix_history = \
                        tracker.apply(persons, frame_resized,
                                      bg_subtraction_resized, number_frame)
                    del persons
                    trayectos = trayectos_ if trayectos_ else trayectos

                    if SAVE_POSITIONS_TO_FILE:
                        if number_frame >= 50:
                            positions_to_file += positions_in_frame

                        for ((x1, y1), (x2, y2)) in rectangles_in_frame:
                            # Draw in green candidate blobs
                            cv2.rectangle(frame_resized_copy,
                                          (int(x1), int(y1)),
                                          (int(x2), int(y2)),
                                          (0, 255, 0), 1)

                        p_matrix_history += frame_p_matrix_history

                    if len(comparisons_by_color_image_aux) > 0:
                        comparisons_by_color_image = \
                            comparisons_by_color_image_aux

                    aux_time = time.time() - t0
                    if number_frame > 200:
                        t_time += aux_time
                        max_t_time = max(aux_time, max_t_time)

                    t0 = time.time()

                    # ################################################# ##
                    # ## COMMUNICATION WITH PATTERN MASTER AND OTHERS # ##
                    # ################################################# ##

                    if number_frame % FPS_OVER_2 == 0:
                        for info in info_to_send:
                            info['tracker_id'] = identifier

                            frame_resized_marks = frame_resized.copy()
                            cv2.rectangle(
                                frame_resized_marks, info['rectangle'][0],
                                info['rectangle'][1], (200, 0, 0), -1)
                            frame_resized_marks = \
                                cv2.addWeighted(frame_resized_marks, 0.2,
                                                frame_resized, 0.8, 0)
                            cv2.circle(frame_resized_marks,
                                       (int(info['last_position'][0]),
                                        int(info['last_position'][1])),
                                       70, (200, 200, 0), -1)
                            frame_resized_marks = \
                                cv2.addWeighted(frame_resized_marks, 0.2,
                                                frame_resized, 0.8, 0)
                            info['img'] = \
                                frame2base64png(frame_resized_marks).decode()
                        # Send info to the pattern recognition
                        # every half second
                        communicator.apply(json.dumps(info_to_send),
                                           routing_key='track_info')

                    if number_frame % (FPS*10) == 0:
                        # Renew the config in pattern recognition every
                        # 10 seconds
                        send_patternrecognition_config(
                            communicator, identifier, patternmaster_conf,
                            resolution_multiplier)

                    aux_time = time.time() - t0
                    if number_frame > 200:
                        pattern_recogn_time += aux_time
                        max_pattern_recogn_time = \
                            max(aux_time, max_pattern_recogn_time)

                t0 = time.time()

                now = dt.now()
                for tracklet in tracklets.values():
                    if getattr(tracklet, 'last_rule', None):
                        time_pass = now - getattr(tracklet, 'last_rule_time')
                        if time_pass.seconds < 9:
                            if SHOW_VIDEO_OUTPUT:
                                cv2.putText(
                                    to_show, tracklet.last_rule,
                                    (int(tracklet.last_point[0]),
                                     int(tracklet.last_point[1])),
                                    font, 0.3 - (time_pass.seconds/30),
                                    (255, 0, 0), 1)
                        else:
                            tracklet.last_rule = None

                if SHOW_VIDEO_OUTPUT:
                    # Draw the journeys of the tracked persons
                    draw_journeys(trayectos, [frame_resized_copy, to_show])

                aux_time = time.time() - t0
                if number_frame > 200:
                    show_info_time += aux_time
                    max_show_info_time = max(aux_time, max_show_info_time)

                if SAVE_POSITIONS_TO_FILE:
                    if number_frame >= 50:
                        persons_in_scene += str(number_frame) + ", " + \
                            str(cant_personas) + ", " + \
                            str(len(trayectos)) + ", " + str(round(
                                (len(trayectos) * .85) +
                                (cant_personas * .15))) + "\n"

                if SHOW_VIDEO_OUTPUT:
                    # #################### ##
                    # ## DISPLAY RESULTS # ##
                    # #################### ##

                    t0 = time.time()

                    big_frame = \
                        np.vstack((np.hstack((bg_subtraction, to_show)),
                                   np.hstack((frame_resized,
                                              frame_resized_copy))))
                    # TEXT INFORMATION
                    # Write FPS in the frame to show

                    cv2.putText(big_frame, 'Current persons detected: ' +
                                str(cant_personas), (20, 20), font, .5,
                                (255, 255, 0), 1)
                    cv2.putText(big_frame, 'Current tracklets: ' +
                                str(len(trayectos)), (20, 40), font, .5,
                                (255, 255, 0), 1)
                    interpol_cant_persons = round(
                        ((len(trayectos) * .7) + (cant_personas * .3)) * .35 +
                        interpol_cant_persons_prev * .65)
                    interpol_cant_persons_prev = interpol_cant_persons
                    cv2.putText(big_frame,
                                'Current tracklets/persons interpol. num: ' +
                                str(round((len(trayectos) * .85) +
                                          (cant_personas * .15))),
                                (20, 60), font, .5, (255, 255, 0), 1)
                    cv2.putText(big_frame, 'FPS: ' + _fps, (20, 80), font, .5,
                                (255, 255, 0), 1)

                    big_frame = cv2.resize(big_frame, (work_w*4, work_h*4))
                    cv2.imshow('result', big_frame)

                    if SHOW_COMPARISONS_BY_COLOR:
                        if len(comparisons_by_color_image) > 0:
                            cv2.imshow('comparisons by color',
                                       comparisons_by_color_image)

                    aux_time = time.time() - t0
                    if number_frame > 200:
                        display_time += aux_time
                        max_display_time = max(aux_time, max_display_time)

                    t0 = time.time()

                    if cv2.waitKey(1) & 0xFF in (ord('q'), ord('Q')):
                        exit_cause = 'CLOSED BY PRESSING "Q|q"'
                        break

                    aux_time = time.time() - t0
                    if number_frame > 200:
                        wait_key_time += aux_time
                        max_wait_key_time = max(aux_time, max_wait_key_time)
                if VERBOSE:
                    print("frame: ", str(number_frame),
                          "; fps: ", str(_fps))

                aux_time = time.time() - t_total
                if number_frame > 200:
                    total_time += aux_time
                    max_total_time = max(aux_time, max_total_time)

        global kill_reader
        kill_reader = True

        cv2.destroyAllWindows()

        if USE_HISTOGRAMS_FOR_PERSON_DETECTION and CREATE_MODEL:
            person_detection.save_histogram()

        number_frame_skip_first = number_frame - 200

        avg_times_text = "Average times::::"
        read_time /= number_frame_skip_first
        avg_times_text += "\nRead time " + str(read_time)
        bg_sub_time /= number_frame_skip_first
        avg_times_text += "\nBackground subtraction time " + str(bg_sub_time)
        blob_det_time /= number_frame_skip_first
        avg_times_text += "\nBlob detector time " + str(blob_det_time)
        person_detection_time /= number_frame_skip_first
        avg_times_text += "\nPerson detector time " + \
                          str(person_detection_time)
        t_time /= number_frame_skip_first
        avg_times_text += "\nTracker time " + str(t_time)
        pattern_recogn_time /= number_frame_skip_first
        avg_times_text += "\nCommunication with pattern recognition time " + \
                          str(pattern_recogn_time)
        show_info_time /= number_frame_skip_first
        avg_times_text += "\nText and paths time " + str(show_info_time)
        display_time /= number_frame_skip_first
        avg_times_text += "\nDisplay time " + str(display_time)
        wait_key_time /= number_frame_skip_first
        avg_times_text += "\ncv2.waitKey time " + str(wait_key_time)
        total_time /= number_frame_skip_first
        avg_times_text += "\nTotal time " + str(total_time)

        avg_times_text += "\n\n\nMax times::::"
        avg_times_text += "\nRead time " + str(max_read_time)
        avg_times_text += "\nBackground subtraction time " + \
                          str(max_bg_sub_time)
        avg_times_text += "\nBlob detector time " + str(max_blob_det_time)
        avg_times_text += "\nPerson detector time " + \
            str(max_person_detection_time)
        avg_times_text += "\nTracker time " + str(max_t_time)
        avg_times_text += "\nCommunication with pattern recognition time " + \
                          str(max_pattern_recogn_time)
        avg_times_text += "\nText and paths time " + str(max_show_info_time)
        avg_times_text += "\nDisplay time " + str(max_display_time)
        avg_times_text += "\ncv2.waitKey time " + str(max_wait_key_time)
        avg_times_text += "\nTotal time " + str(max_total_time)

        print(avg_times_text)

        if SAVE_POSITIONS_TO_FILE:
            with open("../experimental_analysis/raw_results/" + identifier +
                      "-positions.txt", "w") as text_file:
                print(positions_to_file, file=text_file)
            with open("../experimental_analysis/raw_results/" + identifier +
                      "-times.txt", "w") as text_file:
                print(avg_times_text, file=text_file)
            with open("../experimental_analysis/raw_results/" + identifier +
                      "-counter.txt", "w") as text_file:
                print(persons_in_scene, file=text_file)

            with open("../experimental_analysis/raw_results/" + identifier +
                      "-p_matrix.txt", "w") as text_file:
                print(p_matrix_history, file=text_file)

        comm_info = get_status_info_comm()
        comm_info.send_message(json.dumps(dict(
            info_id="EXIT", id=identifier,
            content="CAUSE: " + exit_cause,
            img=frame2base64png(frame_resized).decode())),
            routing_key='info')
    else:
        print(model_load[1])

    if SHOW_COMPARISONS_BY_COLOR:
        cv2.imwrite("comparisons_by_color.png", comparisons_by_color_image)

    exit()