RGB Encoding & Mono with MobilenetSSD & Depth

This example shows how to configure the depthai video encoder in h.265 format to encode the RGB camera input at Full-HD resolution at 30FPS, and transfers the encoded video over XLINK to the host, saving it to disk as a video file. At the same time, a MobileNetv2SSD network is ran on the frames from right grayscale camera, while the application also displays the depth map produced by both of the grayscale cameras. Note that disparity is used in this case, as it colorizes in a more intuitive way.

Pressing Ctrl+C will stop the recording and then convert it using ffmpeg into an mp4 to make it playable. Note that ffmpeg will need to be installed and runnable for the conversion to mp4 to succeed.

Be careful, this example saves encoded video to your host storage. So if you leave it running, you could fill up your storage on your host.

It’s a combination of RGB Encoding and Mono & MobilenetSSD & Depth.

Similiar samples:

Demo

Setup

Please run the install script to download all required dependencies. Please note that this script must be ran from git context, so you have to download the depthai-python repository first and then run the script

git clone https://github.com/luxonis/depthai-python.git
cd depthai-python/examples
python3 install_requirements.py

For additional information, please follow installation guide

This example also requires MobilenetSDD blob (mobilenet-ssd_openvino_2021.2_6shave.blob file) to work - you can download it from here

Source code

Also available on GitHub

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
#!/usr/bin/env python3

from pathlib import Path
import sys
import cv2
import depthai as dai
import numpy as np

# Get argument first
nnPath = str((Path(__file__).parent / Path('models/mobilenet-ssd_openvino_2021.4_6shave.blob')).resolve().absolute())
if len(sys.argv) > 1:
    nnPath = sys.argv[1]

if not Path(nnPath).exists():
    import sys
    raise FileNotFoundError(f'Required file/s not found, please run "{sys.executable} install_requirements.py"')

# MobilenetSSD label texts
labelMap = ["background", "aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow",
            "diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"]

flipRectified = True

# Create pipeline
pipeline = dai.Pipeline()

# Define sources and outputs
camRgb = pipeline.createColorCamera()
videoEncoder = pipeline.createVideoEncoder()
monoRight = pipeline.createMonoCamera()
monoLeft = pipeline.createMonoCamera()
depth = pipeline.createStereoDepth()
manip = pipeline.createImageManip()
nn = pipeline.createMobileNetDetectionNetwork()

videoOut = pipeline.createXLinkOut()
xoutRight = pipeline.createXLinkOut()
disparityOut = pipeline.createXLinkOut()
manipOut = pipeline.createXLinkOut()
nnOut = pipeline.createXLinkOut()

videoOut.setStreamName('h265')
xoutRight.setStreamName('right')
disparityOut.setStreamName('disparity')
manipOut.setStreamName('manip')
nnOut.setStreamName('nn')

# Properties
camRgb.setBoardSocket(dai.CameraBoardSocket.RGB)
camRgb.setResolution(dai.ColorCameraProperties.SensorResolution.THE_1080_P)
monoRight.setBoardSocket(dai.CameraBoardSocket.RIGHT)
monoRight.setResolution(dai.MonoCameraProperties.SensorResolution.THE_400_P)
monoLeft.setBoardSocket(dai.CameraBoardSocket.LEFT)
monoLeft.setResolution(dai.MonoCameraProperties.SensorResolution.THE_400_P)
videoEncoder.setDefaultProfilePreset(1920, 1080, 30, dai.VideoEncoderProperties.Profile.H265_MAIN)

# Note: the rectified streams are horizontally mirrored by default
depth.initialConfig.setConfidenceThreshold(255)
depth.setRectifyMirrorFrame(False)
depth.setRectifyEdgeFillColor(0) # Black, to better see the cutout

nn.setConfidenceThreshold(0.5)
nn.setBlobPath(nnPath)
nn.setNumInferenceThreads(2)
nn.input.setBlocking(False)

# The NN model expects BGR input. By default ImageManip output type would be same as input (gray in this case)
manip.initialConfig.setFrameType(dai.ImgFrame.Type.BGR888p)
manip.initialConfig.setResize(300, 300)

# Linking
camRgb.video.link(videoEncoder.input)
videoEncoder.bitstream.link(videoOut.input)
monoRight.out.link(xoutRight.input)
monoRight.out.link(depth.right)
monoLeft.out.link(depth.left)
depth.disparity.link(disparityOut.input)
depth.rectifiedRight.link(manip.inputImage)
manip.out.link(nn.input)
manip.out.link(manipOut.input)
nn.out.link(nnOut.input)

# Disparity range is used for normalization
disparityMultiplier = 255 / depth.getMaxDisparity()

# Connect to device and start pipeline
with dai.Device(pipeline) as device:

    queueSize = 8
    qRight = device.getOutputQueue("right", queueSize)
    qDisparity = device.getOutputQueue("disparity", queueSize)
    qManip = device.getOutputQueue("manip", queueSize)
    qDet = device.getOutputQueue("nn", queueSize)
    qRgbEnc = device.getOutputQueue('h265', maxSize=30, blocking=True)

    frame = None
    frameManip = None
    frameDisparity = None
    detections = []
    offsetX = (monoRight.getResolutionWidth() - monoRight.getResolutionHeight()) // 2
    color = (255, 0, 0)
    croppedFrame = np.zeros((monoRight.getResolutionHeight(), monoRight.getResolutionHeight()))

    def frameNorm(frame, bbox):
        normVals = np.full(len(bbox), frame.shape[0])
        normVals[::2] = frame.shape[1]
        return (np.clip(np.array(bbox), 0, 1) * normVals).astype(int)

    videoFile = open('video.h265', 'wb')
    cv2.namedWindow("right", cv2.WINDOW_NORMAL)
    cv2.namedWindow("manip", cv2.WINDOW_NORMAL)

    while True:
        inRight = qRight.tryGet()
        inManip = qManip.tryGet()
        inDet = qDet.tryGet()
        inDisparity = qDisparity.tryGet()

        while qRgbEnc.has():
            qRgbEnc.get().getData().tofile(videoFile)

        if inRight is not None:
            frame = inRight.getCvFrame()

        if inManip is not None:
            frameManip = inManip.getCvFrame()

        if inDisparity is not None:
            # Flip disparity frame, normalize it and apply color map for better visualization
            frameDisparity = inDisparity.getCvFrame()
            if flipRectified:
                frameDisparity = cv2.flip(frameDisparity, 1)
            frameDisparity = (frameDisparity*disparityMultiplier).astype(np.uint8)
            frameDisparity = cv2.applyColorMap(frameDisparity, cv2.COLORMAP_JET)

        if inDet is not None:
            detections = inDet.detections

        if frame is not None:
            for detection in detections:
                bbox = frameNorm(croppedFrame, (detection.xmin, detection.ymin, detection.xmax, detection.ymax))
                bbox[::2] += offsetX
                cv2.putText(frame, labelMap[detection.label], (bbox[0] + 10, bbox[1] + 20), cv2.FONT_HERSHEY_TRIPLEX, 0.5, color)
                cv2.putText(frame, f"{int(detection.confidence * 100)}%", (bbox[0] + 10, bbox[1] + 40), cv2.FONT_HERSHEY_TRIPLEX, 0.5, color)
                cv2.rectangle(frame, (bbox[0], bbox[1]), (bbox[2], bbox[3]), color, 2)
            # Show the right cam frame
            cv2.imshow("right", frame)

        if frameDisparity is not None:
            for detection in detections:
                bbox = frameNorm(croppedFrame, (detection.xmin, detection.ymin, detection.xmax, detection.ymax))
                bbox[::2] += offsetX
                cv2.rectangle(frameDisparity, (bbox[0], bbox[1]), (bbox[2], bbox[3]), color, 2)
                cv2.putText(frameDisparity, labelMap[detection.label], (bbox[0] + 10, bbox[1] + 20), cv2.FONT_HERSHEY_TRIPLEX, 0.5, color)
                cv2.putText(frameDisparity, f"{int(detection.confidence * 100)}%", (bbox[0] + 10, bbox[1] + 40), cv2.FONT_HERSHEY_TRIPLEX, 0.5, color)
            # Show the disparity frame
            cv2.imshow("disparity", frameDisparity)

        if frameManip is not None:
            for detection in detections:
                bbox = frameNorm(frameManip, (detection.xmin, detection.ymin, detection.xmax, detection.ymax))
                cv2.rectangle(frameManip, (bbox[0], bbox[1]), (bbox[2], bbox[3]), color, 2)
                cv2.putText(frameManip, labelMap[detection.label], (bbox[0] + 10, bbox[1] + 20), cv2.FONT_HERSHEY_TRIPLEX, 0.5, color)
                cv2.putText(frameManip, f"{int(detection.confidence * 100)}%", (bbox[0] + 10, bbox[1] + 40), cv2.FONT_HERSHEY_TRIPLEX, 0.5, color)
            # Show the manip frame
            cv2.imshow("manip", frameManip)

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

    print("To view the encoded data, convert the stream file (.h265) into a video file (.mp4) using a command below:")
    print("ffmpeg -framerate 30 -i video.h265 -c copy video.mp4")

Also available on GitHub

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
#include <cstdio>
#include <iostream>

// Inludes common necessary includes for development using depthai library
#include "depthai/depthai.hpp"

// MobilenetSSD label texts
static const std::vector<std::string> labelMap = {"background", "aeroplane", "bicycle",     "bird",  "boat",        "bottle", "bus",
                                                  "car",        "cat",       "chair",       "cow",   "diningtable", "dog",    "horse",
                                                  "motorbike",  "person",    "pottedplant", "sheep", "sofa",        "train",  "tvmonitor"};

static std::atomic<bool> flipRectified{true};

int main(int argc, char** argv) {
    using namespace std;
    // Default blob path provided by Hunter private data download
    // Applicable for easier example usage only
    std::string nnPath(BLOB_PATH);

    // If path to blob specified, use that
    if(argc > 1) {
        nnPath = std::string(argv[1]);
    }

    // Print which blob we are using
    printf("Using blob at path: %s\n", nnPath.c_str());

    // Create pipeline
    dai::Pipeline pipeline;

    // Define sources and outputs
    auto camRgb = pipeline.create<dai::node::ColorCamera>();
    auto videoEncoder = pipeline.create<dai::node::VideoEncoder>();
    auto monoRight = pipeline.create<dai::node::MonoCamera>();
    auto monoLeft = pipeline.create<dai::node::MonoCamera>();
    auto depth = pipeline.create<dai::node::StereoDepth>();
    auto manip = pipeline.create<dai::node::ImageManip>();
    auto nn = pipeline.create<dai::node::MobileNetDetectionNetwork>();

    auto videoOut = pipeline.create<dai::node::XLinkOut>();
    auto xoutRight = pipeline.create<dai::node::XLinkOut>();
    auto disparityOut = pipeline.create<dai::node::XLinkOut>();
    auto manipOut = pipeline.create<dai::node::XLinkOut>();
    auto nnOut = pipeline.create<dai::node::XLinkOut>();

    videoOut->setStreamName("h265");
    xoutRight->setStreamName("right");
    disparityOut->setStreamName("disparity");
    manipOut->setStreamName("manip");
    nnOut->setStreamName("nn");

    // Properties
    camRgb->setBoardSocket(dai::CameraBoardSocket::RGB);
    camRgb->setResolution(dai::ColorCameraProperties::SensorResolution::THE_1080_P);
    monoRight->setBoardSocket(dai::CameraBoardSocket::RIGHT);
    monoRight->setResolution(dai::MonoCameraProperties::SensorResolution::THE_400_P);
    monoLeft->setBoardSocket(dai::CameraBoardSocket::LEFT);
    monoLeft->setResolution(dai::MonoCameraProperties::SensorResolution::THE_400_P);
    videoEncoder->setDefaultProfilePreset(1920, 1080, 30, dai::VideoEncoderProperties::Profile::H265_MAIN);

    // Note: the rectified streams are horizontally mirrored by default
    depth->initialConfig.setConfidenceThreshold(255);
    depth->setRectifyMirrorFrame(false);
    depth->setRectifyEdgeFillColor(0);  // Black, to better see the cutout

    nn->setConfidenceThreshold(0.5);
    nn->setBlobPath(nnPath);
    nn->setNumInferenceThreads(2);
    nn->input.setBlocking(false);

    // The NN model expects BGR input-> By default ImageManip output type would be same as input (gray in this case)
    manip->initialConfig.setFrameType(dai::ImgFrame::Type::BGR888p);
    manip->initialConfig.setResize(300, 300);

    // Linking
    camRgb->video.link(videoEncoder->input);
    videoEncoder->bitstream.link(videoOut->input);
    monoRight->out.link(xoutRight->input);
    monoRight->out.link(depth->right);
    monoLeft->out.link(depth->left);
    depth->disparity.link(disparityOut->input);
    depth->rectifiedRight.link(manip->inputImage);
    manip->out.link(nn->input);
    manip->out.link(manipOut->input);
    nn->out.link(nnOut->input);

    // Disparity range is used for normalization
    float disparityMultiplier = 255 / depth->getMaxDisparity();

    // Connect to device and start pipeline
    dai::Device device(pipeline);

    // Queues
    int queueSize = 8;
    auto qRight = device.getOutputQueue("right", queueSize);
    auto qDisparity = device.getOutputQueue("disparity", queueSize);
    auto qManip = device.getOutputQueue("manip", queueSize);
    auto qDet = device.getOutputQueue("nn", queueSize);
    auto qRgbEnc = device.getOutputQueue("h265", 30, true);

    cv::Mat frame;
    cv::Mat frameManip;
    cv::Mat frameDisparity;
    std::vector<dai::ImgDetection> detections;
    int offsetX = (monoRight->getResolutionWidth() - monoRight->getResolutionHeight()) / 2;
    auto color = cv::Scalar(255, 0, 0);

    auto videoFile = std::ofstream("video.h265", std::ios::binary);
    cv::namedWindow("right", cv::WINDOW_NORMAL);
    cv::namedWindow("manip", cv::WINDOW_NORMAL);

    while(true) {
        auto inRight = qRight->tryGet<dai::ImgFrame>();
        auto inManip = qManip->tryGet<dai::ImgFrame>();
        auto inDet = qDet->tryGet<dai::ImgDetections>();
        auto inDisparity = qDisparity->tryGet<dai::ImgFrame>();

        auto out1 = qRgbEnc->get<dai::ImgFrame>();
        videoFile.write((char*)out1->getData().data(), out1->getData().size());

        if(inRight) {
            frame = inRight->getCvFrame();
        }

        if(inManip) {
            frameManip = inManip->getCvFrame();
        }

        if(inDisparity) {
            frameDisparity = inDisparity->getCvFrame();
            if(flipRectified) {
                cv::flip(frameDisparity, frameDisparity, 1);
            }
            frameDisparity.convertTo(frameDisparity, CV_8UC1, disparityMultiplier);
            cv::applyColorMap(frameDisparity, frameDisparity, cv::COLORMAP_JET);
        }

        if(inDet) {
            detections = inDet->detections;
        }

        if(!frame.empty()) {
            for(auto& detection : detections) {
                int x1 = detection.xmin * monoRight->getResolutionHeight() + offsetX;
                int y1 = detection.ymin * monoRight->getResolutionHeight();
                int x2 = detection.xmax * monoRight->getResolutionHeight() + offsetX;
                int y2 = detection.ymax * monoRight->getResolutionHeight();

                int labelIndex = detection.label;
                std::string labelStr = to_string(labelIndex);
                if(labelIndex < labelMap.size()) {
                    labelStr = labelMap[labelIndex];
                }
                cv::putText(frame, labelStr, cv::Point(x1 + 10, y1 + 20), cv::FONT_HERSHEY_TRIPLEX, 0.5, color);
                std::stringstream confStr;
                confStr << std::fixed << std::setprecision(2) << detection.confidence * 100;
                cv::putText(frame, confStr.str(), cv::Point(x1 + 10, y1 + 40), cv::FONT_HERSHEY_TRIPLEX, 0.5, color);
                cv::rectangle(frame, cv::Rect(cv::Point(x1, y1), cv::Point(x2, y2)), color, cv::FONT_HERSHEY_SIMPLEX);
            }
            // Show the right cam frame
            cv::imshow("right", frame);
        }

        if(!frameDisparity.empty()) {
            for(auto& detection : detections) {
                int x1 = detection.xmin * monoRight->getResolutionHeight() + offsetX;
                int y1 = detection.ymin * monoRight->getResolutionHeight();
                int x2 = detection.xmax * monoRight->getResolutionHeight() + offsetX;
                int y2 = detection.ymax * monoRight->getResolutionHeight();

                int labelIndex = detection.label;
                std::string labelStr = to_string(labelIndex);
                if(labelIndex < labelMap.size()) {
                    labelStr = labelMap[labelIndex];
                }
                cv::putText(frameDisparity, labelStr, cv::Point(x1 + 10, y1 + 20), cv::FONT_HERSHEY_TRIPLEX, 0.5, color);
                std::stringstream confStr;
                confStr << std::fixed << std::setprecision(2) << detection.confidence * 100;
                cv::putText(frameDisparity, confStr.str(), cv::Point(x1 + 10, y1 + 40), cv::FONT_HERSHEY_TRIPLEX, 0.5, color);
                cv::rectangle(frameDisparity, cv::Rect(cv::Point(x1, y1), cv::Point(x2, y2)), color, cv::FONT_HERSHEY_SIMPLEX);
            }
            // Show the disparity frame
            cv::imshow("disparity", frameDisparity);
        }

        if(!frameManip.empty()) {
            for(auto& detection : detections) {
                int x1 = detection.xmin * frameManip.cols;
                int y1 = detection.ymin * frameManip.rows;
                int x2 = detection.xmax * frameManip.cols;
                int y2 = detection.ymax * frameManip.rows;

                int labelIndex = detection.label;
                std::string labelStr = to_string(labelIndex);
                if(labelIndex < labelMap.size()) {
                    labelStr = labelMap[labelIndex];
                }
                cv::putText(frameManip, labelStr, cv::Point(x1 + 10, y1 + 20), cv::FONT_HERSHEY_TRIPLEX, 0.5, color);
                std::stringstream confStr;
                confStr << std::fixed << std::setprecision(2) << detection.confidence * 100;
                cv::putText(frameManip, confStr.str(), cv::Point(x1 + 10, y1 + 40), cv::FONT_HERSHEY_TRIPLEX, 0.5, color);
                cv::rectangle(frameManip, cv::Rect(cv::Point(x1, y1), cv::Point(x2, y2)), color, cv::FONT_HERSHEY_SIMPLEX);
            }
            // Show the manip frame
            cv::imshow("manip", frameManip);
        }

        int key = cv::waitKey(1);
        if(key == 'q' || key == 'Q') {
            break;
        }
    }
    cout << "To view the encoded data, convert the stream file (.h265) into a video file (.mp4), using a command below:" << endl;
    cout << "ffmpeg -framerate 30 -i video.h265 -c copy video.mp4" << endl;
    return 0;
}

Got questions?

We’re always happy to help with code or other questions you might have.