Perform stereo depth estimation using the depthai API on images from the host computer in Python.

depthai_stereo_from_host_images.py

import depthai as dai
import cv2
import numpy as np

from imread_from_url import imread_from_url

def add_host_depth(pipeline):

    monoLeftNode = pipeline.create(dai.node.XLinkIn)
    monoRightNode = pipeline.create(dai.node.XLinkIn)

    depthNode = create_depth_node(pipeline)
    monoLeftNode.out.link(depthNode.left)
    monoRightNode.out.link(depthNode.right)
    monoLeftNode.setStreamName('inLeft')
    monoRightNode.setStreamName('inRight')

    xinStereoDepthConfig = pipeline.create(dai.node.XLinkIn)
    xinStereoDepthConfig.out.link(depthNode.inputConfig)
    xinStereoDepthConfig.setStreamName("stereoDepthConfig")

    xoutStereoCfg = pipeline.create(dai.node.XLinkOut)
    depthNode.outConfig.link(xoutStereoCfg.input)
    xoutStereoCfg.setStreamName('stereoConfig')

def create_depth_node(pipeline):
    depthNode = pipeline.create(dai.node.StereoDepth)
    depthNode.setDefaultProfilePreset(dai.node.StereoDepth.PresetMode.HIGH_DENSITY)
    depthNode.setRectifyEdgeFillColor(0)  # Black, to better see the cutout
    depthNode.initialConfig.setMedianFilter(dai.MedianFilter.KERNEL_7x7)
    depthNode.setLeftRightCheck(True)
    depthNode.setExtendedDisparity(True)
    depthNode.setRectification(False)
    depthNode.setInputResolution(width, height)

    depthOutNode = pipeline.create(dai.node.XLinkOut)
    depthNode.disparity.link(depthOutNode.input)
    depthOutNode.setStreamName("disparity")

    return depthNode
def img_to_daiFrame(img, side=dai.CameraBoardSocket.RIGHT):
    if img.shape[2] > 1:
        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

    data = cv2.resize(img, (width, height), interpolation = cv2.INTER_AREA)
    data = data.reshape(height*width)

    imgFrame = dai.ImgFrame()
    imgFrame.setData(data)
    imgFrame.setInstanceNum(side)
    imgFrame.setType(dai.ImgFrame.Type.RAW8)
    imgFrame.setWidth(width)
    imgFrame.setHeight(height)
    return imgFrame


if __name__ == '__main__':

    height, width = 800, 1280

    # Create pipeline
    pipeline = dai.Pipeline()
    add_host_depth(pipeline)

    with dai.Device(pipeline) as device:

        stereoDepthConfigInQueue = device.getInputQueue("stereoDepthConfig")
        currentConfig = device.getOutputQueue("stereoConfig", 8, blocking=False)
        qLeft = device.getInputQueue("inLeft")
        qRight = device.getInputQueue("inRight")
        qDepth = device.getOutputQueue(name="disparity", maxSize=4, blocking=False)

        left_img = imread_from_url("https://vision.middlebury.edu/stereo/data/scenes2003/newdata/cones/im2.png")
        right_img = imread_from_url("https://vision.middlebury.edu/stereo/data/scenes2003/newdata/cones/im6.png")

        imgFrameLeft = img_to_daiFrame(left_img, dai.CameraBoardSocket.LEFT)
        imgFrameRight = img_to_daiFrame(right_img, dai.CameraBoardSocket.RIGHT)

        qLeft.send(imgFrameLeft)
        qRight.send(imgFrameRight)

        config = currentConfig.get()
        maxDisp = config.getMaxDisparity()
        print(f"Max disparity: {maxDisp}")

        disp_map = qDepth.get().getCvFrame()
        color_disp = cv2.applyColorMap((disp_map / maxDisp * 255).astype(np.uint8), cv2.COLORMAP_JET)

        cv2.imshow("disparity", color_disp)
        cv2.waitKey(0)