1)Open CV のインストール

参考ページ：Raspberry Pi 4 に OpenCV を楽に入れる
まず、OpenCVをインストールしてUSBカメラを使用できる状態にします。

依存関係を導入

ターミナルを起動し、下記のコマンドで環境を構築

LXterminal

sudo apt-get install libjpeg-dev libtiff5-dev libjasper-dev libpng-dev
sudo apt-get install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev
sudo apt-get install libxvidcore-dev libx264-dev
sudo apt-get install libatlas-base-dev gfortran
sudo apt-get install libhdf5-dev libhdf5-serial-dev libhdf5-103
sudo apt-get install libqtgui4 libqtwebkit4 libqt4-test python3-pyqt5
sudo apt-get install python3-dev

pipコマンドでOpen CV をインストール

LXterminal

pip3 install opencv-contrib-python==4.1.0.25

これで、画像関係の設定は完了！

2)Coral を設定

coral のホームページを参考に環境を導入
https://coral.ai/docs/edgetpu/api-intro/

LXterminal

echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
sudo apt-get update

sudo apt-get install libedgetpu1-std

設定をインストール

LXterminal

wget https://dl.google.com/coral/edgetpu_api/edgetpu_api_latest.tar.gz -O edgetpu_api.tar.gz --trust-server-names
tar xzf edgetpu_api.tar.gz
cd edgetpu_api
bash ./install.sh

ここでエラーが頻発するときも・・・
bashコマンドでエラーが発生。原因不明。（別の方法は（3）で）
（ご存じでしたら原因を教えていただけると助かります。）

あとはapt-getでもedgetpuをインストールしておく。

LXterminal

sudo apt-get install python3-edgetpu

3)Sampleスクリプトの実行

まずはgithubからsampleをダウンロード

LXterminal

git clone https://github.com/leswright1977/RPi4-Google-Coral

このフォルダ内の_src_の中にわかりやすいサンプルスクリプトがある。

・(2)でbashエラーが発生したかたはsrc内にあるinstall.shを「edgetpu_api」にコピーすると動きました。
（原因は不明です。。。）

あとはUSBカメラとCoralを接続し、sampleを走らせるだけでObject detectionができる。

気になる部分があればコメントをお願いします。
（都度、修正します。）

python_sample

import cv2
import time
import numpy as np
from multiprocessing import Process
from multiprocessing import Queue

import edgetpu.detection.engine
from edgetpu.utils import image_processing
from PIL import Image

# Misc vars
font = cv2.FONT_HERSHEY_SIMPLEX
queuepulls = 0.0
detections = 0
fps = 0.0
qfps = 0.0
confThreshold = 0.6

# init video
cap = cv2.VideoCapture(0)
print("[info] W, H, FPS")
print(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
print(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
print(cap.get(cv2.CAP_PROP_FPS))

frameWidth = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
frameHeight = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))

labels_file = 'tpu_models/coco_labels.txt'
# Read labels from text files. Note, there are missing items from the coco_labels txt hence this!
labels = [None] * 10
with open(labels_file, 'r') as f:
    lines = f.readlines()
for line in lines:
    parts = line.strip().split(maxsplit=1)
    labels.insert(int(parts[0]),str(parts[1])) 
print(labels)

# define the function that handles our processing thread
def classify_frame(img, inputQueue, outputQueue):
    engine = edgetpu.detection.engine.DetectionEngine(\
    'tpu_models/mobilenet_ssd_v2_coco_quant_postprocess_edgetpu.tflite')
    # keep looping
    while True:
        # check to see if there is a frame in our input queue
        if not inputQueue.empty():
            # grab the frame from the input queue
            img = inputQueue.get()
            results = engine.DetectWithImage(img, threshold=0.4,\
            keep_aspect_ratio=True, relative_coord=False, top_k=10)
            data_out = []

            if results:
                for obj in results:
                    inference = []
                    box = obj.bounding_box.flatten().tolist()
                    xmin = int(box[0])
                    ymin = int(box[1])
                    xmax = int(box[2])
                    ymax = int(box[3])

                    inference.extend((obj.label_id,obj.score,xmin,ymin,xmax,ymax))
                    data_out.append(inference)
            outputQueue.put(data_out)
# initialize the input queue (frames), output queue (out),
# and the list of actual detections returned by the child process
inputQueue = Queue(maxsize=1)
outputQueue = Queue(maxsize=1)
img = None
out = None

# construct a child process *indepedent* from our main process of
# execution
print("[INFO] starting process...")
p = Process(target=classify_frame, args=(img,inputQueue,outputQueue,))
p.daemon = True
p.start()

print("[INFO] starting capture...")

# time the frame rate....
timer1 = time.time()
frames = 0
queuepulls = 0
timer2 = 0
t2secs = 0

while(cap.isOpened()):
    # Capture frame-by-frame
    ret, frame = cap.read()
    if ret == True:
        if queuepulls ==1:
            timer2 = time.time()
        # Capture frame-by-frame
        #frame = frame.array
        img = Image.fromarray(frame)
        # if the input queue *is* empty, give the current frame to
        # classify
        if inputQueue.empty():
            inputQueue.put(img)
        # if the output queue *is not* empty, grab the detections
        if not outputQueue.empty():
            out = outputQueue.get()
        if out is not None:
            # loop over the detections
            for detection in out:
                objID = detection[0]
                labeltxt = labels[objID]
                confidence = detection[1]
                xmin = detection[2]
                ymin = detection[3]
                xmax = detection[4]
                ymax = detection[5]
                if confidence > confThreshold:
                    #bounding box
                    cv2.rectangle(frame, (xmin, ymin), (xmax, ymax), color=(0, 255, 255))
                    #label
                    labLen = len(labeltxt)*5+40
                    cv2.rectangle(frame, (xmin-1, ymin-1),\
                    (xmin+labLen, ymin-10), (0,255,255), -1)
                    #labeltext
                    cv2.putText(frame,' '+labeltxt+' '+str(round(confidence,2)),\
                    (xmin,ymin-2), font, 0.3,(0,0,0),1,cv2.LINE_AA)
                    detections +=1 #positive detections    
            queuepulls += 1
        # Display the resulting frame
        cv2.rectangle(frame, (0,0),\
        (frameWidth,20), (0,0,0), -1)

        cv2.rectangle(frame, (0,frameHeight-20),\
        (frameWidth,frameHeight), (0,0,0), -1)
        cv2.putText(frame,'Threshold: '+str(round(confThreshold,1)), (10, 10),\
        cv2.FONT_HERSHEY_SIMPLEX, 0.3,(0, 255, 255), 1, cv2.LINE_AA)

        cv2.putText(frame,'VID FPS: '+str(fps), (frameWidth-80, 10),\
        cv2.FONT_HERSHEY_SIMPLEX, 0.3,(0, 255, 255), 1, cv2.LINE_AA)

        cv2.putText(frame,'TPU FPS: '+str(qfps), (frameWidth-80, 20),\
        cv2.FONT_HERSHEY_SIMPLEX, 0.3,(0, 255, 255), 1, cv2.LINE_AA)

        cv2.putText(frame,'Positive detections: '+str(detections), (10, frameHeight-10),\
        cv2.FONT_HERSHEY_SIMPLEX, 0.3,(0, 255, 255), 1, cv2.LINE_AA)

        cv2.putText(frame,'Elapsed time: '+str(round(t2secs,2)), (150, frameHeight-10),\
        cv2.FONT_HERSHEY_SIMPLEX, 0.3,(0, 255, 255), 1, cv2.LINE_AA)
        

        cv2.namedWindow('Coral',cv2.WINDOW_NORMAL)
        #cv2.resizeWindow('Coral',frameWidth,frameHeight)
        cv2.imshow('Coral',frame)
        
        # FPS calculation
        
        frames += 1
        if frames >= 1:
            end1 = time.time()
            t1secs = end1-timer1
            fps = round(frames/t1secs,2)
        if queuepulls > 1:
            end2 = time.time()
            t2secs = end2-timer2
            qfps = round(queuepulls/t2secs,2)
        

        keyPress = cv2.waitKey(1) & 0xFF #Altering waitkey val can alter the framreate for vid files.
        if keyPress == ord('q'):
            break
        if keyPress == ord('r'):
            confThreshold += 0.1
        if keyPress == ord('t'):
            confThreshold -= 0.1
        if confThreshold >1:
            confThreshold = 1
        if confThreshold <0.4:
            confThreshold = 0.4
    # Break the loop
    else: 
        break
# Everything done, release the vid
cap.release()

cv2.destroyAllWindows()