Live Object Detection with OpenVINO™

This tutorial is also available as a Jupyter notebook that can be cloned directly from GitHub. See the installation guide for instructions to run this tutorial locally on Windows, Linux or macOS.


This notebook demonstrates live object detection with OpenVINO, using the SSDLite MobileNetV2 from Open Model Zoo. Final part of this notebook shows live inference results from a webcam. Additionally, you can also upload a video file.

NOTE: To use this notebook with a webcam, you need to run the notebook on a computer with a webcam. If you run the notebook on a server, the webcam will not work. However, you can still do inference on a video.


import collections
import sys
import tarfile
import time
from pathlib import Path

import cv2
import numpy as np
from IPython import display
from openvino import runtime as ov
from import tf as ov_tf_front

import notebook_utils as utils

The Model

Download the Model

Use the download_file, a function from the notebook_utils file. It automatically creates a directory structure and downloads the selected model. This step is skipped if the package is already downloaded and unpacked. The chosen model comes from the public directory, which means it must be converted into OpenVINO Intermediate Representation (OpenVINO IR).

NOTE: Using a model other than ssdlite_mobilenet_v2 may require different conversion parameters as well as pre- and post-processing.

# A directory where the model will be downloaded.
base_model_dir = Path("model")

# The name of the model from Open Model Zoo
model_name = "ssdlite_mobilenet_v2"

archive_name = Path(f"{model_name}_coco_2018_05_09.tar.gz")
model_url = f"{model_name}/{archive_name}"

# Download the archive
downloaded_model_path = base_model_dir / archive_name
if not downloaded_model_path.exists():
    utils.download_file(model_url,, downloaded_model_path.parent)

# Unpack the model
tf_model_path = base_model_dir / archive_name.with_suffix("").stem / "frozen_inference_graph.pb"
if not tf_model_path.exists():
    with as file:
model/ssdlite_mobilenet_v2_coco_2018_05_09.tar.gz:   0%|          | 0.00/48.7M [00:00<?, ?B/s]

Convert the Model

The pre-trained model is in TensorFlow format. To use it with OpenVINO, convert it to OpenVINO IR format. Use Model Optimizer (mo), a command-line tool from the openvino-dev package. You can achieve the same if you use Python API for Model Optimizer. If the model has been already converted, this step is skipped.

precision = "FP16"
# The output path for the conversion.
converted_model_path = Path("model") / f"{model_name}_{precision.lower()}.xml"

# Convert it to IR if not previously converted
trans_config_path = Path(ov_tf_front.__file__).parent / "ssd_v2_support.json"
if not converted_model_path.exists():
    convert_command = f"mo " \
                      f"--input_model {tf_model_path} " \
                      f"--output_dir {base_model_dir} " \
                      f"--model_name {model_name}_{precision.lower()} " \
                      f"--compress_to_fp16 {True if precision == 'FP16' else False} " \
                      f"--transformations_config={trans_config_path} " \
                      f"--tensorflow_object_detection_api_pipeline_config {tf_model_path.parent}/pipeline.config " \
    ! $convert_command
[ WARNING ]  The Preprocessor block has been removed. Only nodes performing mean value subtraction and scaling (if applicable) are kept.
Check for a new version of Intel(R) Distribution of OpenVINO(TM) toolkit here or on
[ INFO ] The model was converted to IR v11, the latest model format that corresponds to the source DL framework input/output format. While IR v11 is backwards compatible with OpenVINO Inference Engine API v1.0, please use API v2.0 (as of 2022.1) to take advantage of the latest improvements in IR v11.
Find more information about API v2.0 and IR v11 at
[ SUCCESS ] Generated IR version 11 model.
[ SUCCESS ] XML file: /opt/home/k8sworker/cibuilds/ov-notebook/OVNotebookOps-408/.workspace/scm/ov-notebook/notebooks/401-object-detection-webcam/model/ssdlite_mobilenet_v2_fp16.xml
[ SUCCESS ] BIN file: /opt/home/k8sworker/cibuilds/ov-notebook/OVNotebookOps-408/.workspace/scm/ov-notebook/notebooks/401-object-detection-webcam/model/ssdlite_mobilenet_v2_fp16.bin

Load the Model

Only a few lines of code are required to run the model. First, initialize OpenVINO Runtime. Then, read the network architecture and model weights from the .bin and .xml files to compile for the desired device. If you choose GPU you need to wait for a while, as the startup time is much longer than in the case of CPU.

There is a possibility to let OpenVINO decide which hardware offers the best performance. For that purpose, just use AUTO. Remember that for most cases the best hardware is GPU (better performance, but longer startup time).

# Initialize OpenVINO Runtime.
ie_core = ov.Core()
# Read the network and corresponding weights from a file.
model = ie_core.read_model(model=converted_model_path)
# Compile the model for CPU (you can choose manually CPU, GPU, MYRIAD etc.)
# or let the engine choose the best available device (AUTO).
compiled_model = ie_core.compile_model(model=model, device_name="CPU")

# Get the input and output nodes.
input_layer = compiled_model.input(0)
output_layer = compiled_model.output(0)

# Get the input size.
height, width = list(input_layer.shape)[1:3]

Input and output layers have the names of the input node and output node respectively. In the case of SSDLite MobileNetV2, there is 1 input and 1 output.

input_layer.any_name, output_layer.any_name
('image_tensor:0', 'detection_boxes:0')


Process Results

First, list all available classes and create colors for them. Then, in the post-process stage, transform boxes with normalized coordinates [0, 1] into boxes with pixel coordinates [0, image_size_in_px]. Afterward, use non-maximum suppression to reject overlapping detections and those below the probability threshold (0.5). Finally, draw boxes and labels inside them.

classes = [
    "background", "person", "bicycle", "car", "motorcycle", "airplane", "bus", "train",
    "truck", "boat", "traffic light", "fire hydrant", "street sign", "stop sign",
    "parking meter", "bench", "bird", "cat", "dog", "horse", "sheep", "cow", "elephant",
    "bear", "zebra", "giraffe", "hat", "backpack", "umbrella", "shoe", "eye glasses",
    "handbag", "tie", "suitcase", "frisbee", "skis", "snowboard", "sports ball", "kite",
    "baseball bat", "baseball glove", "skateboard", "surfboard", "tennis racket", "bottle",
    "plate", "wine glass", "cup", "fork", "knife", "spoon", "bowl", "banana", "apple",
    "sandwich", "orange", "broccoli", "carrot", "hot dog", "pizza", "donut", "cake", "chair",
    "couch", "potted plant", "bed", "mirror", "dining table", "window", "desk", "toilet",
    "door", "tv", "laptop", "mouse", "remote", "keyboard", "cell phone", "microwave", "oven",
    "toaster", "sink", "refrigerator", "blender", "book", "clock", "vase", "scissors",
    "teddy bear", "hair drier", "toothbrush", "hair brush"

# Colors for the classes above (Rainbow Color Map).
colors = cv2.applyColorMap(
    src=np.arange(0, 255, 255 / len(classes), dtype=np.float32).astype(np.uint8),

def process_results(frame, results, thresh=0.6):
    # The size of the original frame.
    h, w = frame.shape[:2]
    # The 'results' variable is a [1, 1, 100, 7] tensor.
    results = results.squeeze()
    boxes = []
    labels = []
    scores = []
    for _, label, score, xmin, ymin, xmax, ymax in results:
        # Create a box with pixels coordinates from the box with normalized coordinates [0,1].
            tuple(map(int, (xmin * w, ymin * h, (xmax - xmin) * w, (ymax - ymin) * h)))

    # Apply non-maximum suppression to get rid of many overlapping entities.
    # See
    # This algorithm returns indices of objects to keep.
    indices = cv2.dnn.NMSBoxes(
        bboxes=boxes, scores=scores, score_threshold=thresh, nms_threshold=0.6

    # If there are no boxes.
    if len(indices) == 0:
        return []

    # Filter detected objects.
    return [(labels[idx], scores[idx], boxes[idx]) for idx in indices.flatten()]

def draw_boxes(frame, boxes):
    for label, score, box in boxes:
        # Choose color for the label.
        color = tuple(map(int, colors[label]))
        # Draw a box.
        x2 = box[0] + box[2]
        y2 = box[1] + box[3]
        cv2.rectangle(img=frame, pt1=box[:2], pt2=(x2, y2), color=color, thickness=3)

        # Draw a label name inside the box.
            text=f"{classes[label]} {score:.2f}",
            org=(box[0] + 10, box[1] + 30),
            fontScale=frame.shape[1] / 1000,

    return frame

Main Processing Function

Run object detection on the specified source. Either a webcam or a video file.

# Main processing function to run object detection.
def run_object_detection(source=0, flip=False, use_popup=False, skip_first_frames=0):
    player = None
        # Create a video player to play with target fps.
        player = utils.VideoPlayer(
            source=source, flip=flip, fps=30, skip_first_frames=skip_first_frames
        # Start capturing.
        if use_popup:
            title = "Press ESC to Exit"
                winname=title, flags=cv2.WINDOW_GUI_NORMAL | cv2.WINDOW_AUTOSIZE

        processing_times = collections.deque()
        while True:
            # Grab the frame.
            frame =
            if frame is None:
                print("Source ended")
            # If the frame is larger than full HD, reduce size to improve the performance.
            scale = 1280 / max(frame.shape)
            if scale < 1:
                frame = cv2.resize(

            # Resize the image and change dims to fit neural network input.
            input_img = cv2.resize(
                src=frame, dsize=(width, height), interpolation=cv2.INTER_AREA
            # Create a batch of images (size = 1).
            input_img = input_img[np.newaxis, ...]

            # Measure processing time.

            start_time = time.time()
            # Get the results.
            results = compiled_model([input_img])[output_layer]
            stop_time = time.time()
            # Get poses from network results.
            boxes = process_results(frame=frame, results=results)

            # Draw boxes on a frame.
            frame = draw_boxes(frame=frame, boxes=boxes)

            processing_times.append(stop_time - start_time)
            # Use processing times from last 200 frames.
            if len(processing_times) > 200:

            _, f_width = frame.shape[:2]
            # Mean processing time [ms].
            processing_time = np.mean(processing_times) * 1000
            fps = 1000 / processing_time
                text=f"Inference time: {processing_time:.1f}ms ({fps:.1f} FPS)",
                org=(20, 40),
                fontScale=f_width / 1000,
                color=(0, 0, 255),

            # Use this workaround if there is flickering.
            if use_popup:
                cv2.imshow(winname=title, mat=frame)
                key = cv2.waitKey(1)
                # escape = 27
                if key == 27:
                # Encode numpy array to jpg.
                _, encoded_img = cv2.imencode(
                    ext=".jpg", img=frame, params=[cv2.IMWRITE_JPEG_QUALITY, 100]
                # Create an IPython image.
                i = display.Image(data=encoded_img)
                # Display the image in this notebook.
    # ctrl-c
    except KeyboardInterrupt:
    # any different error
    except RuntimeError as e:
        if player is not None:
            # Stop capturing.
        if use_popup:


Run Live Object Detection

Use a webcam as the video input. By default, the primary webcam is set with source=0. If you have multiple webcams, each one will be assigned a consecutive number starting at 0. Set flip=True when using a front-facing camera. Some web browsers, especially Mozilla Firefox, may cause flickering. If you experience flickering, set use_popup=True.

NOTE: To use this notebook with a webcam, you need to run the notebook on a computer with a webcam. If you run the notebook on a server (for example, Binder), the webcam will not work. Popup mode may not work if you run this notebook on a remote computer (for example, Binder).

Run the object detection:

run_object_detection(source=0, flip=True, use_popup=False)
Cannot open camera 0
[ WARN:0@45.104] global cap_v4l.cpp:982 open VIDEOIO(V4L2:/dev/video0): can't open camera by index
[ERROR:0@45.104] global obsensor_uvc_stream_channel.cpp:156 getStreamChannelGroup Camera index out of range

Run Object Detection on a Video File

If you do not have a webcam, you can still run this demo with a video file. Any format supported by OpenCV will work.

video_file = "../data/video/Coco Walking in Berkeley.mp4"

run_object_detection(source=video_file, flip=False, use_popup=False)
Source ended