Stereo Depth from host

This example shows depth map from host using stereo images. There are 3 depth modes which you can select inside the code:

  1. lr_check: used for better occlusion handling. For more information click here

  2. extended_disparity: suitable for short range objects. For more information click here

  3. subpixel: suitable for long range. For more information click here

Otherwise a median with kernel_7x7 is activated.

Similiar samples:

Setup

Please run the following command to install the required dependencies

 python3 -m pip install -U pip
 python3 -m pip install opencv-python
 python3 -m pip install -U --force-reinstall depthai

For additional information, please follow installation guide

This example also requires dataset folder - 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
178
179
180
181
182
183
184
185
#!/usr/bin/env python3

import cv2
import numpy as np
import depthai as dai
from time import sleep
import datetime
import argparse
from pathlib import Path

datasetDefault = str((Path(__file__).parent / Path('models/dataset')).resolve().absolute())
parser = argparse.ArgumentParser()
parser.add_argument('-dataset', nargs='?', help="Path to recorded frames", default=datasetDefault)
args = parser.parse_args()

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


# StereoDepth config options.
out_depth = False  # Disparity by default
out_rectified = True   # Output and display rectified streams
lrcheck = True   # Better handling for occlusions
extended = False  # Closer-in minimum depth, disparity range is doubled
subpixel = True   # Better accuracy for longer distance, fractional disparity 32-levels
median = dai.StereoDepthProperties.MedianFilter.KERNEL_7x7

# Sanitize some incompatible options
if lrcheck or extended or subpixel:
    median = dai.StereoDepthProperties.MedianFilter.MEDIAN_OFF

print("StereoDepth config options: ")
print("Left-Right check: ", lrcheck)
print("Extended disparity: ", extended)
print("Subpixel: ", subpixel)
print("Median filtering: ", median)

right_intrinsic = [[860.0, 0.0, 640.0], [0.0, 860.0, 360.0], [0.0, 0.0, 1.0]]

def create_stereo_depth_pipeline():
    print("Creating Stereo Depth pipeline: ", end='')

    print("XLINK IN -> STEREO -> XLINK OUT")
    pipeline = dai.Pipeline()

    monoLeft = pipeline.createXLinkIn()
    monoRight = pipeline.createXLinkIn()
    stereo = pipeline.createStereoDepth()
    xoutLeft = pipeline.createXLinkOut()
    xoutRight = pipeline.createXLinkOut()
    xoutDepth = pipeline.createXLinkOut()
    xoutDisparity = pipeline.createXLinkOut()
    xoutRectifLeft = pipeline.createXLinkOut()
    xoutRectifRight = pipeline.createXLinkOut()

    monoLeft.setStreamName('in_left')
    monoRight.setStreamName('in_right')

    stereo.setConfidenceThreshold(200)
    stereo.setRectifyEdgeFillColor(0) # Black, to better see the cutout
    stereo.setMedianFilter(median) # KERNEL_7x7 default
    stereo.setLeftRightCheck(lrcheck)
    stereo.setExtendedDisparity(extended)
    stereo.setSubpixel(subpixel)

    stereo.setEmptyCalibration() # Set if the input frames are already rectified
    stereo.setInputResolution(1280, 720)

    xoutLeft.setStreamName('left')
    xoutRight.setStreamName('right')
    xoutDepth.setStreamName('depth')
    xoutDisparity.setStreamName('disparity')
    xoutRectifLeft.setStreamName('rectified_left')
    xoutRectifRight.setStreamName('rectified_right')

    monoLeft.out.link(stereo.left)
    monoRight.out.link(stereo.right)
    stereo.syncedLeft.link(xoutLeft.input)
    stereo.syncedRight.link(xoutRight.input)
    if out_depth:
        stereo.depth.link(xoutDepth.input)
    stereo.disparity.link(xoutDisparity.input)
    if out_rectified:
        stereo.rectifiedLeft.link(xoutRectifLeft.input)
        stereo.rectifiedRight.link(xoutRectifRight.input)

    streams = ['left', 'right']
    if out_rectified:
        streams.extend(['rectified_left', 'rectified_right'])
    streams.extend(['disparity', 'depth'])

    return pipeline, streams

def convert_to_cv2_frame(name, image):
    baseline = 75 #mm
    focal = right_intrinsic[0][0]
    max_disp = 96
    disp_type = np.uint8
    disp_levels = 1
    if (extended):
        max_disp *= 2
    if (subpixel):
        max_disp *= 32
        disp_type = np.uint16
        disp_levels = 32

    data, w, h = image.getData(), image.getWidth(), image.getHeight()
    if name == 'depth':
        # this contains FP16 with (lrcheck or extended or subpixel)
        frame = np.array(data).astype(np.uint8).view(np.uint16).reshape((h, w))
    elif name == 'disparity':
        disp = np.array(data).astype(np.uint8).view(disp_type).reshape((h, w))

        # Compute depth from disparity
        with np.errstate(divide='ignore'):
            depth = (disp_levels * baseline * focal / disp).astype(np.uint16)

        if 1: # Optionally, extend disparity range to better visualize it
            frame = (disp * 255. / max_disp).astype(np.uint8)

        if 1: # Optionally, apply a color map
            frame = cv2.applyColorMap(frame, cv2.COLORMAP_HOT)

    else: # mono streams / single channel
        frame = np.array(data).reshape((h, w)).astype(np.uint8)
        if name.startswith('rectified_'):
            frame = cv2.flip(frame, 1)
        if name == 'rectified_right':
            last_rectif_right = frame
    return frame

pipeline, streams = create_stereo_depth_pipeline()

print("Connecting and starting the pipeline")
# Connect to device and start pipeline
with dai.Device(pipeline) as device:

    inStreams = ['in_right', 'in_left']
    inStreamsCameraID = [dai.CameraBoardSocket.RIGHT, dai.CameraBoardSocket.LEFT]
    in_q_list = []
    for s in inStreams:
        q = device.getInputQueue(s)
        in_q_list.append(q)

    # Create a receive queue for each stream
    q_list = []
    for s in streams:
        q = device.getOutputQueue(s, 8, blocking=False)
        q_list.append(q)

    # Need to set a timestamp for input frames, for the sync stage in Stereo node
    timestamp_ms = 0
    index = 0
    while True:
        # Handle input streams, if any
        if in_q_list:
            dataset_size = 2  # Number of image pairs
            frame_interval_ms = 500
            for i, q in enumerate(in_q_list):
                path = args.dataset + '/' + str(index) + '/' + q.getName() + '.png'
                data = cv2.imread(path, cv2.IMREAD_GRAYSCALE).reshape(720*1280)
                tstamp = datetime.timedelta(seconds = timestamp_ms // 1000,
                                            milliseconds = timestamp_ms % 1000)
                img = dai.ImgFrame()
                img.setData(data)
                img.setTimestamp(tstamp)
                img.setInstanceNum(inStreamsCameraID[i])
                img.setType(dai.ImgFrame.Type.RAW8)
                img.setWidth(1280)
                img.setHeight(720)
                q.send(img)
                if timestamp_ms == 0:  # Send twice for first iteration
                    q.send(img)
                print("Sent frame: {:25s}".format(path), 'timestamp_ms:', timestamp_ms)
            timestamp_ms += frame_interval_ms
            index = (index + 1) % dataset_size
            sleep(frame_interval_ms / 1000)
        # Handle output streams
        for q in q_list:
            if q.getName() in ['left', 'right', 'depth']: continue
            frame = convert_to_cv2_frame(q.getName(), q.get())
            cv2.imshow(q.getName(), frame)
        if cv2.waitKey(1) == ord('q'):
            break

Got questions?

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