環(huán)境

• 首先安裝一下matplotlib庫：
  sudo pip install matplotlib

• 下載1.4.1的tensorflow
  https://github.com/lhelontra/tensorflow-on-arm/releases

• 安裝

sudo pip uninstall tensorflow
sudo pip install --upgrade tensorflow-1.4.1-cp27-none-linux_armv7l.whl

準備模型

• 下載tensorflow提供的models API并解壓董朝，我這里解壓后的目錄為models_master，下載路徑：
  https://github.com/tensorflow/models/tree/master/research/object_detection/models
• 下載訓(xùn)練好的模型并放到上一步models_master下的object_detection/models目錄罐柳，下載路徑：
  https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md
  這里下載幾個典型的：ssd_mobilenet_v1_coco_2017_11_17洗鸵、faster_rcnn_resnet101_coco和mask_rcnn_inception_v2_coco
  注：做物體檢測的網(wǎng)絡(luò)有很多種恬惯，如faster rcnn浙宜，ssd，yolo等等蝶棋，通過不同維度的對比嵌赠，各個網(wǎng)絡(luò)都有各自的優(yōu)勢塑荒。
  畢竟樹莓派計算能力有限，我們這里先選擇專門為速度優(yōu)化過最快的網(wǎng)絡(luò)SSD姜挺，以及經(jīng)典的faster-rcnn作對比齿税，再加上能顯示mask的高端網(wǎng)絡(luò)，炊豪，凌箕，
  事實上yolo v3剛出來，比SSD更快溜在，而faster rcnn相對來說運行慢的多了陌知，后面可以都嘗試對比一下，目前先把基線系統(tǒng)搭建好掖肋。

Protobuf 安裝與配置

• 說明
  protobuf是Google開發(fā)的一種混合語言數(shù)據(jù)標(biāo)準仆葡，提供了一種輕便高效的結(jié)構(gòu)化數(shù)據(jù)存儲格式，可以用于結(jié)構(gòu)化數(shù)據(jù)序列化志笼。很適合做數(shù)據(jù)存儲或 RPC 數(shù)據(jù)交換格式沿盅。可用于通訊協(xié)議纫溃、數(shù)據(jù)存儲等領(lǐng)域的語言無關(guān)腰涧、平臺無關(guān)、可擴展的序列化結(jié)構(gòu)數(shù)據(jù)格式紊浩。目前提供了 C++窖铡、Java疗锐、Python 三種語言的 API。
  下載地址：https://github.com/google/protobuf/releases
  我們這里下載最新版本 protobuf-all-3.5.1.tar.gz
• 安裝

tar -xf  protobuf-all-3.5.1.tar.gz  
cd protobuf-3.5.1  
./configure   
make   
make check   ->這一步是檢查編譯是否正確费彼，耗時非常長滑臊，可略過
sudo make install  
sudo ldconfig  ->更新庫搜索路徑，否則可能找不到庫文件

如果運行了make check箍铲，結(jié)果如下雇卷，可以看到所有的測試用例都PASS了，說明編譯正確：

============================================================================
Testsuite summary for Protocol Buffers 3.5.1
============================================================================
# TOTAL: 7
# PASS:  7
# SKIP:  0
# XFAIL: 0
# FAIL:  0
# XPASS: 0
# ERROR: 0
============================================================================

• 配置
  配置的目的是將proto格式的數(shù)據(jù)轉(zhuǎn)換為python格式颠猴，從而可以在python腳本中調(diào)用关划，進入目錄models-master/research，運行：

protoc object_detection/protos/*.proto --python_out=.

轉(zhuǎn)換完畢后可以看到在object_detection/protos/目錄下多了許多*.py文件翘瓮。

代碼

這里的代碼很簡單贮折，因為基本實現(xiàn)都已經(jīng)有了，我們只是調(diào)用一下接口實現(xiàn)功能即可资盅。

import numpy as np
import os
import sys
import tarfile
import tensorflow as tf
import cv2
import time

from collections import defaultdict

# This is needed since the notebook is stored in the object_detection folder.
sys.path.append("../..")

from object_detection.utils import label_map_util
from object_detection.utils import visualization_utils as vis_util

# What model to download.
MODEL_NAME = 'ssd_mobilenet_v1_coco_2017_11_17'
#MODEL_NAME = 'faster_rcnn_resnet101_coco_11_06_2017'
#MODEL_NAME = 'ssd_inception_v2_coco_11_06_2017'
MODEL_FILE = MODEL_NAME + '.tar.gz'

# Path to frozen detection graph. This is the actual model that is used for the object detection.
PATH_TO_CKPT = MODEL_NAME + '/frozen_inference_graph.pb'

# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = os.path.join('/home/yinan/object_detect/models-master/research/object_detection/data', 'mscoco_label_map.pbtxt')

#extract the ssd_mobilenet
start = time.clock()
NUM_CLASSES = 90
#opener = urllib.request.URLopener()
#opener.retrieve(DOWNLOAD_BASE + MODEL_FILE, MODEL_FILE)
tar_file = tarfile.open(MODEL_FILE)
for file in tar_file.getmembers():
   file_name = os.path.basename(file.name)
   if 'frozen_inference_graph.pb' in file_name:
      tar_file.extract(file, os.getcwd())
end= time.clock()
print('load the model',(end-start))
detection_graph = tf.Graph()
with detection_graph.as_default():
  od_graph_def = tf.GraphDef()
  with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
    serialized_graph = fid.read()
    od_graph_def.ParseFromString(serialized_graph)
    tf.import_graph_def(od_graph_def, name='')

label_map = label_map_util.load_labelmap(PATH_TO_LABELS)

categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
category_index = label_map_util.create_category_index(categories)

cap = cv2.VideoCapture(0)
with detection_graph.as_default():
  with tf.Session(graph=detection_graph) as sess:
      writer = tf.summary.FileWriter("logs/", sess.graph)
      sess.run(tf.global_variables_initializer())
      while(1):
        start = time.clock()
        ret, frame = cap.read()
        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
        image_np=frame
        # the array based representation of the image will be used later in order to prepare the
        # result image with boxes and labels on it.
        # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
        image_np_expanded = np.expand_dims(image_np, axis=0)
        image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
        # Each box represents a part of the image where a particular object was detected.
        boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
        # Each score represent how level of confidence for each of the objects.
        # Score is shown on the result image, together with the class label.
        scores = detection_graph.get_tensor_by_name('detection_scores:0')
        classes = detection_graph.get_tensor_by_name('detection_classes:0')
        num_detections = detection_graph.get_tensor_by_name('num_detections:0')
        # Actual detection.

        (boxes, scores, classes, num_detections) = sess.run(
          [boxes, scores, classes, num_detections],
          feed_dict={image_tensor: image_np_expanded})
        # Visualization of the results of a detection.
        vis_util.visualize_boxes_and_labels_on_image_array(
          image_np,
          np.squeeze(boxes),
          np.squeeze(classes).astype(np.int32),
          np.squeeze(scores),
          category_index,
          use_normalized_coordinates=True,
          line_thickness=6)
        end = time.clock()
        #print('frame:',1.0/(end - start))
        print 'One frame detect take time:',end - start

        cv2.imshow("capture", image_np)
        print('after cv2 show')
        cv2.waitKey(1)
cap.release()
cv2.destroyAllWindows()

保存為 detect.py脱货，到目錄models-master/research/object_detection/models下。

運行

命令：

sudo chmod 666 /dev/video0
python detect.py

效果

SSD模型

下圖可以看到律姨，SSD模型加載模型花了8s，差不多一張圖識別時間在5s：

image.png

PS. 為什么把房間識別成了book...

faster-RCNN模型

faster-RCNN臼疫，加載模型83s择份，內(nèi)存不夠，跑不起來烫堤。荣赶。。

image.png

mask SSD模型

mask模型可以描繪出輪廓鸽斟，看起來更高端拔创，加載模型25s，遇到個問題：

image.png

接下來查一下
CPU占用率100%富蓄，內(nèi)存占用60%多