Video & MobilenetSSD

1#!/usr/bin/env python3
2
3from pathlib import Path
4import sys
5import cv2
6import depthai as dai
7import numpy as np
8from time import monotonic
9
10# Get argument first
11parentDir = Path(__file__).parent
12nnPath = str((parentDir / Path('../models/mobilenet-ssd_openvino_2021.4_8shave.blob')).resolve().absolute())
13videoPath = str((parentDir / Path('../models/construction_vest.mp4')).resolve().absolute())
14if len(sys.argv) > 2:
15    nnPath = sys.argv[1]
16    videoPath = sys.argv[2]
17
18if not Path(nnPath).exists() or not Path(videoPath).exists():
19    import sys
20    raise FileNotFoundError(f'Required file/s not found, please run "{sys.executable} install_requirements.py"')
21
22# MobilenetSSD label texts
23labelMap = ["background", "aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow",
24            "diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"]
25
26# Create pipeline
27pipeline = dai.Pipeline()
28
29# Define sources and outputs
30nn = pipeline.create(dai.node.MobileNetDetectionNetwork)
31
32xinFrame = pipeline.create(dai.node.XLinkIn)
33nnOut = pipeline.create(dai.node.XLinkOut)
34
35xinFrame.setStreamName("inFrame")
36nnOut.setStreamName("nn")
37
38# Properties
39nn.setConfidenceThreshold(0.5)
40nn.setBlobPath(nnPath)
41nn.setNumInferenceThreads(2)
42nn.input.setBlocking(False)
43
44# Linking
45xinFrame.out.link(nn.input)
46nn.out.link(nnOut.input)
47
48# Connect to device and start pipeline
49with dai.Device(pipeline) as device:
50
51    # Input queue will be used to send video frames to the device.
52    qIn = device.getInputQueue(name="inFrame")
53    # Output queue will be used to get nn data from the video frames.
54    qDet = device.getOutputQueue(name="nn", maxSize=4, blocking=False)
55
56    frame = None
57    detections = []
58
59    # nn data, being the bounding box locations, are in <0..1> range - they need to be normalized with frame width/height
60    def frameNorm(frame, bbox):
61        normVals = np.full(len(bbox), frame.shape[0])
62        normVals[::2] = frame.shape[1]
63        return (np.clip(np.array(bbox), 0, 1) * normVals).astype(int)
64
65    def to_planar(arr: np.ndarray, shape: tuple) -> np.ndarray:
66        return cv2.resize(arr, shape).transpose(2, 0, 1).flatten()
67
68    def displayFrame(name, frame):
69        for detection in detections:
70            bbox = frameNorm(frame, (detection.xmin, detection.ymin, detection.xmax, detection.ymax))
71            cv2.putText(frame, labelMap[detection.label], (bbox[0] + 10, bbox[1] + 20), cv2.FONT_HERSHEY_TRIPLEX, 0.5, 255)
72            cv2.putText(frame, f"{int(detection.confidence * 100)}%", (bbox[0] + 10, bbox[1] + 40), cv2.FONT_HERSHEY_TRIPLEX, 0.5, 255)
73            cv2.rectangle(frame, (bbox[0], bbox[1]), (bbox[2], bbox[3]), (255, 0, 0), 2)
74        # Show the frame
75        cv2.imshow(name, frame)
76
77    cap = cv2.VideoCapture(videoPath)
78    while cap.isOpened():
79        read_correctly, frame = cap.read()
80        if not read_correctly:
81            break
82
83        img = dai.ImgFrame()
84        img.setData(to_planar(frame, (300, 300)))
85        img.setTimestamp(monotonic())
86        img.setWidth(300)
87        img.setHeight(300)
88        qIn.send(img)
89
90        inDet = qDet.tryGet()
91
92        if inDet is not None:
93            detections = inDet.detections
94
95        if frame is not None:
96            displayFrame("rgb", frame)
97
98        if cv2.waitKey(1) == ord('q'):
99            break