Mono & MobilenetSSD with spatial data

This example shows how to run MobileNetv2SSD on the rectified right input frame, and how to display both the preview, detections, depth map and spatial information (X,Y,Z). It’s similar to example RGB & MobilenetSSD except it has spatial data. X,Y,Z coordinates are relative to the center of depth map.

setConfidenceThreshold - confidence threshold above which objects are detected

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
173
174
175
176
177
#!/usr/bin/env python3

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

'''
Mobilenet SSD device side decoding demo
  The "mobilenet-ssd" model is a Single-Shot multibox Detection (SSD) network intended
  to perform object detection. This model is implemented using the Caffe* framework.
  For details about this model, check out the repository <https://github.com/chuanqi305/MobileNet-SSD>.
'''

# 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"]

syncNN = True
flipRectified = True

# Create pipeline
pipeline = dai.Pipeline()

# Define sources and outputs
monoLeft = pipeline.createMonoCamera()
monoRight = pipeline.createMonoCamera()
stereo = pipeline.createStereoDepth()
spatialDetectionNetwork = pipeline.createMobileNetSpatialDetectionNetwork()
imageManip = pipeline.createImageManip()

xoutManip = pipeline.createXLinkOut()
nnOut = pipeline.createXLinkOut()
depthRoiMap = pipeline.createXLinkOut()
xoutDepth = pipeline.createXLinkOut()

xoutManip.setStreamName("right")
nnOut.setStreamName("detections")
depthRoiMap.setStreamName("boundingBoxDepthMapping")
xoutDepth.setStreamName("depth")

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

monoLeft.setResolution(dai.MonoCameraProperties.SensorResolution.THE_400_P)
monoLeft.setBoardSocket(dai.CameraBoardSocket.LEFT)
monoRight.setResolution(dai.MonoCameraProperties.SensorResolution.THE_400_P)
monoRight.setBoardSocket(dai.CameraBoardSocket.RIGHT)

# StereoDepth
stereo.initialConfig.setConfidenceThreshold(255)

# Define a neural network that will make predictions based on the source frames
spatialDetectionNetwork.setConfidenceThreshold(0.5)
spatialDetectionNetwork.setBlobPath(nnPath)
spatialDetectionNetwork.input.setBlocking(False)
spatialDetectionNetwork.setBoundingBoxScaleFactor(0.5)
spatialDetectionNetwork.setDepthLowerThreshold(100)
spatialDetectionNetwork.setDepthUpperThreshold(5000)

# Linking
monoLeft.out.link(stereo.left)
monoRight.out.link(stereo.right)

imageManip.out.link(spatialDetectionNetwork.input)
if syncNN:
    spatialDetectionNetwork.passthrough.link(xoutManip.input)
else:
    imageManip.out.link(xoutManip.input)

spatialDetectionNetwork.out.link(nnOut.input)
spatialDetectionNetwork.boundingBoxMapping.link(depthRoiMap.input)

stereo.rectifiedRight.link(imageManip.inputImage)
stereo.depth.link(spatialDetectionNetwork.inputDepth)
spatialDetectionNetwork.passthroughDepth.link(xoutDepth.input)

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

    # Output queues will be used to get the rgb frames and nn data from the outputs defined above
    previewQueue = device.getOutputQueue(name="right", maxSize=4, blocking=False)
    detectionNNQueue = device.getOutputQueue(name="detections", maxSize=4, blocking=False)
    depthRoiMapQueue = device.getOutputQueue(name="boundingBoxDepthMapping", maxSize=4, blocking=False)
    depthQueue = device.getOutputQueue(name="depth", maxSize=4, blocking=False)

    rectifiedRight = None
    detections = []

    startTime = time.monotonic()
    counter = 0
    fps = 0
    color = (255, 255, 255)

    while True:
        inRectified = previewQueue.get()
        inDet = detectionNNQueue.get()
        inDepth = depthQueue.get()

        counter += 1
        currentTime = time.monotonic()
        if (currentTime - startTime) > 1:
            fps = counter / (currentTime - startTime)
            counter = 0
            startTime = currentTime

        rectifiedRight = inRectified.getCvFrame()
        if flipRectified:
            rectifiedRight = cv2.flip(rectifiedRight, 1)

        depthFrame = inDepth.getFrame()
        depthFrameColor = cv2.normalize(depthFrame, None, 255, 0, cv2.NORM_INF, cv2.CV_8UC1)
        depthFrameColor = cv2.equalizeHist(depthFrameColor)
        depthFrameColor = cv2.applyColorMap(depthFrameColor, cv2.COLORMAP_HOT)

        detections = inDet.detections
        if len(detections) != 0:
            boundingBoxMapping = depthRoiMapQueue.get()
            roiDatas = boundingBoxMapping.getConfigData()

            for roiData in roiDatas:
                roi = roiData.roi
                roi = roi.denormalize(depthFrameColor.shape[1], depthFrameColor.shape[0])
                topLeft = roi.topLeft()
                bottomRight = roi.bottomRight()
                xmin = int(topLeft.x)
                ymin = int(topLeft.y)
                xmax = int(bottomRight.x)
                ymax = int(bottomRight.y)
                cv2.rectangle(depthFrameColor, (xmin, ymin), (xmax, ymax), color, cv2.FONT_HERSHEY_SCRIPT_SIMPLEX)

        # If the rectifiedRight is available, draw bounding boxes on it and show the rectifiedRight
        height = rectifiedRight.shape[0]
        width = rectifiedRight.shape[1]
        for detection in detections:
            if flipRectified:
                swap = detection.xmin
                detection.xmin = 1 - detection.xmax
                detection.xmax = 1 - swap
            # Denormalize bounding box
            x1 = int(detection.xmin * width)
            x2 = int(detection.xmax * width)
            y1 = int(detection.ymin * height)
            y2 = int(detection.ymax * height)

            try:
                label = labelMap[detection.label]
            except:
                label = detection.label

            cv2.putText(rectifiedRight, str(label), (x1 + 10, y1 + 20), cv2.FONT_HERSHEY_TRIPLEX, 0.5, 255)
            cv2.putText(rectifiedRight, "{:.2f}".format(detection.confidence*100), (x1 + 10, y1 + 35), cv2.FONT_HERSHEY_TRIPLEX, 0.5, 255)
            cv2.putText(rectifiedRight, f"X: {int(detection.spatialCoordinates.x)} mm", (x1 + 10, y1 + 50), cv2.FONT_HERSHEY_TRIPLEX, 0.5, 255)
            cv2.putText(rectifiedRight, f"Y: {int(detection.spatialCoordinates.y)} mm", (x1 + 10, y1 + 65), cv2.FONT_HERSHEY_TRIPLEX, 0.5, 255)
            cv2.putText(rectifiedRight, f"Z: {int(detection.spatialCoordinates.z)} mm", (x1 + 10, y1 + 80), cv2.FONT_HERSHEY_TRIPLEX, 0.5, 255)

            cv2.rectangle(rectifiedRight, (x1, y1), (x2, y2), color, cv2.FONT_HERSHEY_SIMPLEX)

        cv2.putText(rectifiedRight, "NN fps: {:.2f}".format(fps), (2, rectifiedRight.shape[0] - 4), cv2.FONT_HERSHEY_TRIPLEX, 0.4, color)
        cv2.imshow("depth", depthFrameColor)
        cv2.imshow("rectified right", rectifiedRight)

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

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
#include <chrono>
#include <iostream>

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

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

int main(int argc, char** argv) {
    using namespace std;
    using namespace std::chrono;
    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 monoLeft = pipeline.create<dai::node::MonoCamera>();
    auto monoRight = pipeline.create<dai::node::MonoCamera>();
    auto stereo = pipeline.create<dai::node::StereoDepth>();
    auto spatialDetectionNetwork = pipeline.create<dai::node::MobileNetSpatialDetectionNetwork>();
    auto imageManip = pipeline.create<dai::node::ImageManip>();

    auto xoutManip = pipeline.create<dai::node::XLinkOut>();
    auto nnOut = pipeline.create<dai::node::XLinkOut>();
    auto depthRoiMap = pipeline.create<dai::node::XLinkOut>();
    auto xoutDepth = pipeline.create<dai::node::XLinkOut>();

    xoutManip->setStreamName("right");
    nnOut->setStreamName("detections");
    depthRoiMap->setStreamName("boundingBoxDepthMapping");
    xoutDepth->setStreamName("depth");

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

    monoLeft->setResolution(dai::MonoCameraProperties::SensorResolution::THE_400_P);
    monoLeft->setBoardSocket(dai::CameraBoardSocket::LEFT);
    monoRight->setResolution(dai::MonoCameraProperties::SensorResolution::THE_400_P);
    monoRight->setBoardSocket(dai::CameraBoardSocket::RIGHT);

    // StereoDepth
    stereo->initialConfig.setConfidenceThreshold(255);

    // Define a neural network that will make predictions based on the source frames
    spatialDetectionNetwork->setConfidenceThreshold(0.5f);
    spatialDetectionNetwork->setBlobPath(nnPath);
    spatialDetectionNetwork->input.setBlocking(false);
    spatialDetectionNetwork->setBoundingBoxScaleFactor(0.5);
    spatialDetectionNetwork->setDepthLowerThreshold(100);
    spatialDetectionNetwork->setDepthUpperThreshold(5000);

    // Linking
    monoLeft->out.link(stereo->left);
    monoRight->out.link(stereo->right);

    imageManip->out.link(spatialDetectionNetwork->input);
    if(syncNN) {
        spatialDetectionNetwork->passthrough.link(xoutManip->input);
    } else {
        imageManip->out.link(xoutManip->input);
    }

    spatialDetectionNetwork->out.link(nnOut->input);
    spatialDetectionNetwork->boundingBoxMapping.link(depthRoiMap->input);

    stereo->rectifiedRight.link(imageManip->inputImage);
    stereo->depth.link(spatialDetectionNetwork->inputDepth);
    spatialDetectionNetwork->passthroughDepth.link(xoutDepth->input);

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

    // Output queues will be used to get the rgb frames and nn data from the outputs defined above
    auto previewQueue = device.getOutputQueue("right", 4, false);
    auto detectionNNQueue = device.getOutputQueue("detections", 4, false);
    auto depthRoiMapQueue = device.getOutputQueue("boundingBoxDepthMapping", 4, false);
    auto depthQueue = device.getOutputQueue("depth", 4, false);

    auto startTime = steady_clock::now();
    int counter = 0;
    float fps = 0;
    auto color = cv::Scalar(255, 255, 255);

    while(true) {
        auto inRectified = previewQueue->get<dai::ImgFrame>();
        auto inDet = detectionNNQueue->get<dai::SpatialImgDetections>();
        auto inDepth = depthQueue->get<dai::ImgFrame>();

        counter++;
        auto currentTime = steady_clock::now();
        auto elapsed = duration_cast<duration<float>>(currentTime - startTime);
        if(elapsed > seconds(1)) {
            fps = counter / elapsed.count();
            counter = 0;
            startTime = currentTime;
        }

        cv::Mat rectifiedRight = inRectified->getCvFrame();
        if(flipRectified) cv::flip(rectifiedRight, rectifiedRight, 1);

        cv::Mat depthFrame = inDepth->getFrame();
        cv::Mat depthFrameColor;
        cv::normalize(depthFrame, depthFrameColor, 255, 0, cv::NORM_INF, CV_8UC1);
        cv::equalizeHist(depthFrameColor, depthFrameColor);
        cv::applyColorMap(depthFrameColor, depthFrameColor, cv::COLORMAP_HOT);

        auto detections = inDet->detections;
        if(!detections.empty()) {
            auto boundingBoxMapping = depthRoiMapQueue->get<dai::SpatialLocationCalculatorConfig>();
            auto roiDatas = boundingBoxMapping->getConfigData();

            for(auto roiData : roiDatas) {
                auto roi = roiData.roi;
                roi = roi.denormalize(depthFrameColor.cols, depthFrameColor.rows);
                auto topLeft = roi.topLeft();
                auto bottomRight = roi.bottomRight();
                auto xmin = (int)topLeft.x;
                auto ymin = (int)topLeft.y;
                auto xmax = (int)bottomRight.x;
                auto ymax = (int)bottomRight.y;
                cv::rectangle(depthFrameColor, cv::Rect(cv::Point(xmin, ymin), cv::Point(xmax, ymax)), color, cv::FONT_HERSHEY_SIMPLEX);
            }
        }

        for(auto& detection : detections) {
            if(flipRectified) {
                auto swap = detection.xmin;
                detection.xmin = 1 - detection.xmax;
                detection.xmax = 1 - swap;
            }
            int x1 = detection.xmin * rectifiedRight.cols;
            int y1 = detection.ymin * rectifiedRight.rows;
            int x2 = detection.xmax * rectifiedRight.cols;
            int y2 = detection.ymax * rectifiedRight.rows;

            int labelIndex = detection.label;
            std::string labelStr = to_string(labelIndex);
            if(labelIndex < labelMap.size()) {
                labelStr = labelMap[labelIndex];
            }
            cv::putText(rectifiedRight, labelStr, cv::Point(x1 + 10, y1 + 20), cv::FONT_HERSHEY_TRIPLEX, 0.5, 255);
            std::stringstream confStr;
            confStr << std::fixed << std::setprecision(2) << detection.confidence * 100;
            cv::putText(rectifiedRight, confStr.str(), cv::Point(x1 + 10, y1 + 35), cv::FONT_HERSHEY_TRIPLEX, 0.5, 255);

            std::stringstream depthX;
            depthX << "X: " << (int)detection.spatialCoordinates.x << " mm";
            cv::putText(rectifiedRight, depthX.str(), cv::Point(x1 + 10, y1 + 50), cv::FONT_HERSHEY_TRIPLEX, 0.5, 255);
            std::stringstream depthY;
            depthY << "Y: " << (int)detection.spatialCoordinates.y << " mm";
            cv::putText(rectifiedRight, depthY.str(), cv::Point(x1 + 10, y1 + 65), cv::FONT_HERSHEY_TRIPLEX, 0.5, 255);
            std::stringstream depthZ;
            depthZ << "Z: " << (int)detection.spatialCoordinates.z << " mm";
            cv::putText(rectifiedRight, depthZ.str(), cv::Point(x1 + 10, y1 + 80), cv::FONT_HERSHEY_TRIPLEX, 0.5, 255);

            cv::rectangle(rectifiedRight, cv::Rect(cv::Point(x1, y1), cv::Point(x2, y2)), color, cv::FONT_HERSHEY_SIMPLEX);
        }

        std::stringstream fpsStr;
        fpsStr << std::fixed << std::setprecision(2) << fps;
        cv::putText(rectifiedRight, fpsStr.str(), cv::Point(2, rectifiedRight.rows - 4), cv::FONT_HERSHEY_TRIPLEX, 0.4, color);

        cv::imshow("depth", depthFrameColor);
        cv::imshow("rectified right", rectifiedRight);

        int key = cv::waitKey(1);
        if(key == 'q' || key == 'Q') {
            return 0;
        }
    }
    return 0;
}

Got questions?

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