Tensorflow Object Detection API in WebCam

tf_objdet_api_webcam.py

import sys

sys.path.append("/opt/tf_model/research")
sys.path.append("/opt/tf_model/research/object_detection")

import os
import cv2
import numpy as np
import tensorflow as tf
from utils import label_map_util
from utils import visualization_utils as vis_util

if tf.__version__ < '1.4.0':
    raise ImportError('Please upgrade your tensorflow installation to v1.4.* or later!')

print("OpenCV version :  {0}".format(cv2.__version__))

# Should run under docker container from tensorflow_object_detection
ROOT = '/opt/tf_model/research/object_detection/'

# Download pre-train SSD-MobileNet model from
# http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_coco_2017_11_17.tar.gz

MODEL_ROOT = '/datasets/tf_model/'
MODEL_NAME = 'ssd_mobilenet_v1_coco_2017_11_17'
PATH_TO_CKPT = os.path.join(MODEL_ROOT, MODEL_NAME, 'frozen_inference_graph.pb')

# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = os.path.join(ROOT, 'data', 'mscoco_label_map.pbtxt')
NUM_CLASSES = 90


def detect_objects(image_np, sess, detection_graph):
    # 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')

    t1 = cv2.getTickCount()
    # Actual detection.
    (boxes, scores, classes, num_detections) = sess.run(
        [boxes, scores, classes, num_detections],
        feed_dict={image_tensor: image_np_expanded})
    t2 = cv2.getTickCount()
    print((t2 - t1) / cv2.getTickFrequency())

    # 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)
    return image_np


if __name__ == '__main__':
    # This is needed since the notebook is stored in the object_detection folder.

    video_capture = cv2.VideoCapture(0)
    if not video_capture.isOpened():
        print('No video camera found')
        exit()

    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)

    with detection_graph.as_default():
        with tf.Session(graph=detection_graph) as sess:
            # Definite input and output Tensors for detection_graph
            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.
            detection_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.
            detection_scores = detection_graph.get_tensor_by_name('detection_scores:0')
            detection_classes = detection_graph.get_tensor_by_name('detection_classes:0')
            num_detections = detection_graph.get_tensor_by_name('num_detections:0')

            while True:
                ret, frame = video_capture.read()
                frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
                result_rgb = detect_objects(frame_rgb, sess, detection_graph)

                result_bgr = cv2.cvtColor(result_rgb, cv2.COLOR_RGB2BGR)

                cv2.imshow('Video', result_bgr)

                if cv2.waitKey(1) & 0xFF == ord('q'):
                    break

    video_capture.release()
    cv2.destroyAllWindows()