23 - Auto Exposure on ROI

This example shows how to dynamically set the Auto Exposure (AE) of the RGB camera dynamically, during application runtime, based on bounding box position

Demo

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 MobilenetSDD blob (mobilenet-ssd_openvino_2021.2_5shave.blob file) to work - you can download it from here

Usage

By default, AutoExposure region is adjusted based on neural network output. If desired, the region can be set manually. You can do so by pressing one of the following buttons:

  • w - move AE region up

  • s - move AE region down

  • a - move AE region left

  • d - move AE region right

  • n - deactivate manual region (switch back to nn-based roi)

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

# Press WASD to move a manual ROI window for auto-exposure control.
# Press N to go back to the region controlled by the NN detections.

# Get argument first
nnPath = str((Path(__file__).parent / Path('models/mobilenet-ssd_openvino_2021.2_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"')

previewSize = (300, 300)

# Start defining a pipeline
pipeline = dai.Pipeline()

# Define a source - color camera
camRgb = pipeline.createColorCamera()
camRgb.setPreviewSize(*previewSize)
camRgb.setInterleaved(False)

camControlIn = pipeline.createXLinkIn()
camControlIn.setStreamName('camControl')
camControlIn.out.link(camRgb.inputControl)

# Define a neural network that will make predictions based on the source frames
nn = pipeline.createMobileNetDetectionNetwork()
nn.setConfidenceThreshold(0.5)
nn.setBlobPath(nnPath)
nn.setNumInferenceThreads(2)
nn.input.setBlocking(False)
camRgb.preview.link(nn.input)

# Create outputs
xoutRgb = pipeline.createXLinkOut()
xoutRgb.setStreamName("rgb")
camRgb.preview.link(xoutRgb.input)

nnOut = pipeline.createXLinkOut()
nnOut.setStreamName("nn")
nn.out.link(nnOut.input)

# 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"]


def clamp(num, v0, v1):
    return max(v0, min(num, v1))


def asControl(roi):
    camControl = dai.CameraControl()
    camControl.setAutoExposureRegion(*roi)
    return camControl


class AutoExposureRegion:
    step = 10
    position = (0, 0)
    size = (100, 100)
    resolution = camRgb.getResolutionSize()
    maxDims = previewSize[0], previewSize[1]

    def grow(self, x=0, y=0):
        self.size = (
            clamp(x + self.size[0], 1, self.maxDims[0]),
            clamp(y + self.size[1], 1, self.maxDims[1])
        )

    def move(self, x=0, y=0):
        self.position = (
            clamp(x + self.position[0], 0, self.maxDims[0]),
            clamp(y + self.position[1], 0, self.maxDims[1])
        )

    def endPosition(self):
        return (
            clamp(self.position[0] + self.size[0], 0, self.maxDims[0]),
            clamp(self.position[1] + self.size[1], 0, self.maxDims[1]),
        )

    def toRoi(self):
        roi = np.array([*self.position, *self.size])
        # Convert to absolute camera coordinates
        roi = roi * self.resolution[1] // 300
        roi[0] += (self.resolution[0] - self.resolution[1]) // 2  # x offset for device crop
        return roi

    @staticmethod
    def bboxToRoi(bbox):
        startX, startY = bbox[:2]
        width, height = bbox[2] - startX, bbox[3] - startY
        roi = frameNorm(np.empty(camRgb.getResolutionSize()), (startX, startY, width, height))
        return roi


# Connect and start the 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
    qControl = device.getInputQueue(name="camControl")
    qRgb = device.getOutputQueue(name="rgb", maxSize=4, blocking=False)
    qDet = device.getOutputQueue(name="nn", maxSize=4, blocking=False)
    frame = None
    detections = []

    nnRegion = True
    region = AutoExposureRegion()

    # nn data (bounding box locations) are in <0..1> range - they need to be normalized with frame width/height
    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)

    def displayFrame(name, frame):
        for detection in detections:
            bbox = frameNorm(frame, (detection.xmin, detection.ymin, detection.xmax, detection.ymax))
            cv2.rectangle(frame, (bbox[0], bbox[1]), (bbox[2], bbox[3]), (255, 0, 0), 2)
            cv2.putText(frame, labelMap[detection.label], (bbox[0] + 10, bbox[1] + 20), cv2.FONT_HERSHEY_TRIPLEX, 0.5, 255)
            cv2.putText(frame, f"{int(detection.confidence * 100)}%", (bbox[0] + 10, bbox[1] + 40), cv2.FONT_HERSHEY_TRIPLEX, 0.5, 255)
        if not nnRegion:
            cv2.rectangle(frame, region.position, region.endPosition(), (0, 255, 0), 2)
        cv2.imshow(name, frame)

    while True:
        # Instead of get (blocking), we use tryGet (nonblocking) which will return the available data or None otherwise
        inRgb = qRgb.tryGet()
        inDet = qDet.tryGet()

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

        if inDet is not None:
            detections = inDet.detections

            if nnRegion and len(detections) > 0:
                bbox = (detections[0].xmin, detections[0].ymin, detections[0].xmax, detections[0].ymax)
                qControl.send(asControl(AutoExposureRegion.bboxToRoi(bbox)))

        if frame is not None:
            displayFrame("rgb", frame)

        key = cv2.waitKey(1)
        if key == ord('n'):
            print("AE ROI controlled by NN")
            nnRegion = True
        elif key in [ord('w'), ord('a'), ord('s'), ord('d'), ord('+'), ord('-')]:
            nnRegion = False
            if key == ord('a'):
                region.move(x=-region.step)
            if key == ord('d'):
                region.move(x=region.step)
            if key == ord('w'):
                region.move(y=-region.step)
            if key == ord('s'):
                region.move(y=region.step)
            if key == ord('+'):
                region.grow(x=10, y=10)
                region.step = region.step + 1
            if key == ord('-'):
                region.grow(x=-10, y=-10)
                region.step = max(region.step - 1, 1)
            print(f"Setting static AE ROI: {region.toRoi()} (on frame: {[*region.position, *region.endPosition()]})")
            qControl.send(asControl(region.toRoi()))
        elif key == ord('q'):
            break

Got questions?

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