OAK相机如何获取3D坐标？

编辑：OAK中国
首发：oakchina.cn
喜欢的话，请多多👍⭐️✍

▌前言

Hello，大家好，这里是OAK中国，我是助手君。

我们都知道OAK相机是板载AI+CV的深度相机，能够做3D测距，而大家经常好奇的问题就是：OAK相机如何获取3D坐标？

具体的方式你可以参考这个示例

import math
import numpy as np
import depthai as dai

class HostSpatialsCalc:
    # We need device object to get calibration data
    def __init__(self, device):
        calibData = device.readCalibration()
        # Required information for calculating spatial coordinates on the host
        self.monoHFOV = np.deg2rad(calibData.getFov(dai.CameraBoardSocket.LEFT))

        # Values
        self.DELTA = 5
        self.THRESH_LOW = 200 # 20cm
        self.THRESH_HIGH = 30000 # 30m

    def setLowerThreshold(self, threshold_low):
        self.THRESH_LOW = threshold_low
    def setUpperThreshold(self, threshold_low):
        self.THRESH_HIGH = threshold_low
    def setDeltaRoi(self, delta):
        self.DELTA = delta

    def _check_input(self, roi, frame): # Check if input is ROI or point. If point, convert to ROI
        if len(roi) == 4: return roi
        if len(roi) != 2: raise ValueError("You have to pass either ROI (4 values) or point (2 values)!")
        # Limit the point so ROI won't be outside the frame
        self.DELTA = 5 # Take 10x10 depth pixels around point for depth averaging
        x = min(max(roi[0], self.DELTA), frame.shape[1] - self.DELTA)
        y = min(max(roi[1], self.DELTA), frame.shape[0] - self.DELTA)
        return (x-self.DELTA,y-self.DELTA,x+self.DELTA,y+self.DELTA)

    def _calc_angle(self, frame, offset):
        return math.atan(math.tan(self.monoHFOV / 2.0) * offset / (frame.shape[1] / 2.0))

    # roi has to be list of ints
    def calc_spatials(self, depthFrame, roi, averaging_method=np.mean):
        roi = self._check_input(roi, depthFrame) # If point was passed, convert it to ROI
        xmin, ymin, xmax, ymax = roi

        # Calculate the average depth in the ROI.
        depthROI = depthFrame[ymin:ymax, xmin:xmax]
        inRange = (self.THRESH_LOW <= depthROI) & (depthROI <= self.THRESH_HIGH)

        averageDepth = averaging_method(depthROI[inRange])

        centroid = { # Get centroid of the ROI
            'x': int((xmax + xmin) / 2),
            'y': int((ymax + ymin) / 2)
        }

        midW = int(depthFrame.shape[1] / 2) # middle of the depth img width
        midH = int(depthFrame.shape[0] / 2) # middle of the depth img height
        bb_x_pos = centroid['x'] - midW
        bb_y_pos = centroid['y'] - midH

        angle_x = self._calc_angle(depthFrame, bb_x_pos)
        angle_y = self._calc_angle(depthFrame, bb_y_pos)

        spatials = {
            'z': averageDepth,
            'x': averageDepth * math.tan(angle_x),
            'y': -averageDepth * math.tan(angle_y)
        }
        return spatials, centroid

其他相关的示例：

1. StereoDepth node API
2. SpatialLocationCalculator node api
3. SpatialLocationCalculator
4. RGB & TinyYolo with spatial data
5. RGB & MobilenetSSD with spatial data
6. Mono & MobilenetSSD with spatial data

▌参考资料

https://docs.oakchina.cn/en/latest/
https://www.oakchina.cn/selection-guide/

---

OAK中国
| OpenCV AI Kit在中国区的官方代理商和技术服务商
| 追踪AI技术和产品新动态

戳「+关注」获取最新资讯↗↗