【opencv-python-tensorflow object detection API】利用opencv讀取到rtsp的實時幀，采用目标檢測模型對實時視訊流進行檢測（可用！）

• 利用

  tensorflow object detection API

  來訓練适合工作目标檢測模型，其中使用和訓練方式部落格連結如下：

  【Tensorflow object detection API】使用SSD-Mobilenet訓練模型+ubuntu 16.04+python3（步驟十厘清晰！）

• 如下代碼能夠利用opencv實時讀取到rtsp的視訊流，并且采用多線程方式解決了opencv的花屏問題，将視訊流送進目标檢測模型，進行目标檢測：

import threading
# 導入各種包
import numpy as np
import sys
import tensorflow as tf
import time
from distutils.version import StrictVersion

if StrictVersion(tf.__version__) < StrictVersion('1.9.0'):
    raise ImportError('Please upgrade your TensorFlow installation to v1.9.* or later!')
# 增加cv包，以及擷取攝像頭裝置号
import cv2

from utils import label_map_util
from utils import visualization_utils as vis_util

# inhere
PATH_TO_CKPT = '/*/research/object_detection/ssd_model_fpn/model/frozen_inference_graph.pb'

# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = '/*/research/object_detection/ssd_model/pascal_label_map.pbtxt'

NUM_CLASSES = 20

class RTSCapture(cv2.VideoCapture):
    _cur_frame = None
    _reading = False
    schemes = ["rtsp://","rtmp://"]
    @staticmethod
    def create(url, *schemes):
        rtscap = RTSCapture(url)
        rtscap.frame_receiver = threading.Thread(target=rtscap.recv_frame, daemon=True)
        rtscap.schemes.extend(schemes)
        if isinstance(url, str) and url.startswith(tuple(rtscap.schemes)):
            rtscap._reading = True
        elif isinstance(url, int):
            pass
        return rtscap

    def isStarted(self):
        ok = self.isOpened()
        if ok and self._reading:
            ok = self.frame_receiver.is_alive()
        return ok

    def recv_frame(self):
        while self._reading and self.isOpened():
            ok, frame = self.read()
            if not ok: break
            self._cur_frame = frame
        self._reading = False

    def read2(self):
        frame = self._cur_frame
        self._cur_frame = None
        return frame is not None, frame

    def start_read(self):
        self.frame_receiver.start()
        self.read_latest_frame = self.read2 if self._reading else self.read

    def stop_read(self):
        self._reading = False
        if self.frame_receiver.is_alive(): self.frame_receiver.join()


if __name__ == '__main__':
    if len(sys.argv) < 2:
        print("usage:")
        print("need rtsp://xxx")
        sys.exit()

    # load graph
    detection_graph = tf.Graph()
    with detection_graph.as_default():
        od_graph_def = tf.GraphDef()
        with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
            serialized_graph = fid.read()
            od_graph_def.ParseFromString(serialized_graph)
            tf.import_graph_def(od_graph_def, name='')

    label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
    categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES,
                                                                use_display_name=True)
    category_index = label_map_util.create_category_index(categories)

    # open rtsp
    rtscap = RTSCapture.create(sys.argv[1])
    rtscap.start_read()

    with detection_graph.as_default():
        with tf.Session(graph=detection_graph) as sess:
            while rtscap.isStarted():
                ok, image_np = rtscap.read_latest_frame()
                if cv2.waitKey(1) & 0xFF == ord('q'):
                    break
                if not ok:
                    continue
                #time_start = time.time()
                # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
                image_np_expanded = np.expand_dims(image_np, axis=0)
                image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
                # Each box represents a part of the image where a particular object was detected.
                boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
                # Each score represent how level of confidence for each of the objects.
                # Score is shown on the result image, together with the class label.
                scores = detection_graph.get_tensor_by_name('detection_scores:0')
                classes = detection_graph.get_tensor_by_name('detection_classes:0')
                num_detections = detection_graph.get_tensor_by_name('num_detections:0')
                # Actual detection.
                (boxes, scores, classes, num_detections) = sess.run(
                    [boxes, scores, classes, num_detections],
                    feed_dict={image_tensor: image_np_expanded})
                # Visualization of the results of a detection.
                vis_util.visualize_boxes_and_labels_on_image_array(
                    image_np, np.squeeze(boxes),
                    np.squeeze(classes).astype(np.int32),
                    np.squeeze(scores), category_index,
                    use_normalized_coordinates=True,
                    line_thickness=8)

                cv2.imshow('object detection', cv2.resize(image_np, (1500, 800)))
                #time_end = time.time()
                #print('time cost', (time_end - time_start) * 1000, 'ms')


    rtscap.stop_read()
    rtscap.release()
    cv2.destroyAllWindows()