本篇博客應該是全網第一篇將ROS和PaddlePaddle做結合，并且手把手教基于python3部署的文章，目的是為了把ROS和Paddle Inference打通，方便大家基于飛槳做ROS+CV方面的任務！本文只是Demo，大家跑通該Demo后即可將自己的模型替換，實現你自己的創意~

快速體驗

paddle_inference_ros_demo

paddle_inference_ros_demo功能包是基于paddle_inference_ros開發的，幫助開發者快速體驗paddle_inference在ROS環境下的推理部署效果的功能包。可以直接進入該項目鏈接進行查看和快速體驗。

項目地址：

https://gitee.com/irvingao/paddle_inference_ros_demo

1.環境準備

（1）硬件平臺

Jetson nano（Jetson系列開發板配置和以下都相同）

Jetson系列基礎環境配置：

https://blog.csdn.net/qq_45779334/article/details/108611797

（2）軟件環境

• Ubuntu18.04

• python3.6.9

• ROS Melodic

• PaddlePaddle-gpu 2.1.1

PaddlePaddle：跑通Paddle inference Demo：

https://blog.csdn.net/qq_45779334/article/details/118611953

待Paddle inference GPU預測正常之后再將其遷移進ROS中。

2.編譯python3的cv_bridge

在ROS中想使用原生python3的Paddle Inference，最重要的就是需要重新編譯基于python3的cv_bridge，只有我們在編譯完成python3后，才能基于python3實現Paddle Inference目標檢測、分類、分割等相關節點。

所以編譯基于python3的cv_bridge便是最基礎和最重要的一步，該博客詳細介紹了完整的編譯過程，按步驟進行操作即可成功。

ROS——基于Ubuntu18.04和ROS Melodic編譯python3的cv_bridge：

https://blog.csdn.net/qq_45779334/article/details/119641789

3.創建python3Paddle inference ROS節點

這里將以機器人開發中最常用的目標檢測為例，進行部署演示，大家可以自行修改代碼，完成分類、分割等任務的ROS節點。

（1）初始化paddle_ros_ws工作空間

mkdir -p paddle_ros_ws/src && cd paddle_ros_ws/src
catkin_init_workspace

（2）創建功能包

catkin_create_pkg py3_camera rospy rosmsg roscpp
catkin_create_pkg py3_infer rospy rosmsg roscpp

（3）編寫python圖像發布和paddle inference預測節點

①攝像頭發布節點

cd py3_camera && mkdir scripts
cd scripts && touch camera.py
chmod +x camera.py
cd ../..

camera.py

#!/usr/bin/env python3
# coding:utf-8


import cv2
import numpy as np
import rospy
from std_msgs.msg import Header
from sensor_msgs.msg import Image
from cv_bridge import CvBridge , CvBridgeError
import time


if __name__=="__main__":
  import sys 
  print(sys.version) # 查看python版本
  capture = cv2.VideoCapture(0) # 定義攝像頭
  rospy.init_node('camera_node', anonymous=True) #定義節點
  image_pub=rospy.Publisher('/image_view/image_raw', Image, queue_size = 1) #定義話題


  while not rospy.is_shutdown():  # Ctrl C正常退出，如果異常退出會報錯device busy！
    start = time.time()
    ret, frame = capture.read()
    if ret: # 如果有畫面再執行
      # frame = cv2.flip(frame,0)  #垂直鏡像操作
      frame = cv2.flip(frame,1)  #水平鏡像操作  
  
      ros_frame = Image()
      header = Header(stamp = rospy.Time.now())
      header.frame_id = "Camera"
      ros_frame.header=header
      ros_frame.width = 640
      ros_frame.height = 480
      ros_frame.encoding = "bgr8"
      ros_frame.step = 1920
      ros_frame.data = np.array(frame).tostring() #圖片格式轉換
      image_pub.publish(ros_frame) #發布消息
      end = time.time() 
      print("cost time:", end-start ) # 看一下每一幀的執行時間，從而確定合適的rate
      rate = rospy.Rate(25) # 10hz 


  capture.release()
  cv2.destroyAllWindows() 
  print("quit successfully!")

②paddle inference目標檢測節點

cd py3_infer && mkdir scripts
cd scripts && touch pp_infer.py
chmod +x pp_infer.py

pp_infer.py

#!/usr/bin/env python3
# coding:utf-8


import cv2
import numpy as np


import rospy
from sensor_msgs.msg import Image
from cv_bridge import CvBridge, CvBridgeError


from paddle.inference import Config
from paddle.inference import PrecisionType
from paddle.inference import create_predictor
import yaml
import time


# ————————————————圖像預處理函數———————————————— #


def resize(img, target_size):
  """resize to target size"""
  if not isinstance(img, np.ndarray):
    raise TypeError('image type is not numpy.')
  im_shape = img.shape
  im_size_min = np.min(im_shape[0:2])
  im_size_max = np.max(im_shape[0:2])
  im_scale_x = float(target_size) / float(im_shape[1])
  im_scale_y = float(target_size) / float(im_shape[0])
  img = cv2.resize(img, None, None, fx=im_scale_x, fy=im_scale_y)
  return img


def normalize(img, mean, std):
  img = img / 255.0
  mean = np.array(mean)[np.newaxis, np.newaxis, :]
  std = np.array(std)[np.newaxis, np.newaxis, :]
  img -= mean
  img /= std
  return img


def preprocess(img, img_size):
  mean = [0.485, 0.456, 0.406]
  std = [0.229, 0.224, 0.225]
  img = resize(img, img_size)
  resize_img = img
  img = img[:, :, ::-1].astype('float32') # bgr -> rgb
  img = normalize(img, mean, std)
  img = img.transpose((2, 0, 1)) # hwc -> chw
  return img[np.newaxis, :], resize_img


# ——————————————————————模型配置、預測相關函數—————————————————————————— #
def predict_config(model_file, params_file):
  '''
  函數功能：初始化預測模型predictor
  函數輸入：模型結構文件，模型參數文件
  函數輸出：預測器predictor
  '''
  # 根據預測部署的實際情況，設置Config
  config = Config()
  # 讀取模型文件
  config.set_prog_file(model_file)
  config.set_params_file(params_file)
  # Config默認是使用CPU預測，若要使用GPU預測，需要手動開啟，設置運行的GPU卡號和分配的初始顯存。
  config.enable_use_gpu(400, 0)
  # 可以設置開啟IR優化、開啟內存優化。
  config.switch_ir_optim()
  config.enable_memory_optim()
  # config.enable_tensorrt_engine(workspace_size=1 << 30, precision_mode=PrecisionType.Float32,max_batch_size=1, min_subgraph_size=5, use_static=False, use_calib_mode=False)
  predictor = create_predictor(config)
  return predictor


def predict(predictor, img):
  
  '''
  函數功能：初始化預測模型predictor
  函數輸入：模型結構文件，模型參數文件
  函數輸出：預測器predictor
  '''
  input_names = predictor.get_input_names()
  for i, name in enumerate(input_names):
    input_tensor = predictor.get_input_handle(name)
    input_tensor.reshape(img[i].shape)
    input_tensor.copy_from_cpu(img[i].copy())
  # 執行Predictor
  predictor.run()
  # 獲取輸出
  results = []
  # 獲取輸出
  output_names = predictor.get_output_names()
  for i, name in enumerate(output_names):
    output_tensor = predictor.get_output_handle(name)
    output_data = output_tensor.copy_to_cpu()
    results.append(output_data)
  return results


# ——————————————————————后處理函數—————————————————————————— #
def draw_bbox_image(frame, result, label_list, threshold=0.5):
  
  for res in result:
    cat_id, score, bbox = res[0], res[1], res[2:]
    if score < threshold:
     continue
    xmin, ymin, xmax, ymax = bbox
    cv2.rectangle(frame, (int(xmin), int(ymin)), (int(xmax), int(ymax)), (255,0,255), 2)
    label_id = label_list[int(cat_id)]
    print('label is {}, bbox is {}'.format(label_id, bbox))
    try:
      # #cv2.putText(圖像, 文字, (x, y), 字體, 大小, (b, g, r), 寬度)
      cv2.putText(frame, label_id, (int(xmin), int(ymin-2)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255,0,0), 2)
      cv2.putText(frame, str(round(score,2)), (int(xmin-35), int(ymin-2)), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0,255,0), 2)
    except KeyError:
      pass


def callback(data):
  global bridge, predictor, im_size, im_shape, scale_factor, label_list
  cv_img = bridge.imgmsg_to_cv2(data, "bgr8")
  img_data, cv_img = preprocess(cv_img, im_size)    
  # 預測
  result = predict(predictor, [im_shape, img_data, scale_factor])
  draw_bbox_image(cv_img, result[0], label_list, threshold=0.1)
  cv2.imshow("cv_img", cv_img)


  cv2.waitKey(1)


if __name__ == '__main__':
  import sys 
  print(sys.version) # 查看python版本
  
  # 初始化節點
  rospy.init_node('ppinfer_node', anonymous=True)
  bridge = CvBridge()


  # 模型文件路徑(最好寫絕對路徑)
  model_dir = '/home/nano/workspace/paddle_ros_ws/src/py3_infer/scripts/yolov3_r50vd_dcn_270e_coco/'
  # 從infer_cfg.yml中讀出label
  infer_cfg = open(model_dir + 'infer_cfg.yml')
  data = infer_cfg.read()
  yaml_reader = yaml.load(data)
  label_list = yaml_reader['label_list']
  print(label_list)


  # 配置模型參數
  model_file = model_dir + "model.pdmodel"
  params_file = model_dir + "model.pdiparams"
  
  # 圖像尺寸相關參數初始化
  try:
    img = bridge.imgmsg_to_cv2(data, "bgr8")
  except AttributeError:
    img = np.zeros((224,224,3), np.uint8)
  im_size = 224
  scale_factor = np.array([im_size * 1. / img.shape[0], im_size * 1. / img.shape[1]]).reshape((1, 2)).astype(np.float32)
  im_shape = np.array([im_size, im_size]).reshape((1, 2)).astype(np.float32)


  # 初始化預測模型
  predictor = predict_config(model_file, params_file)


  rospy.Subscriber('/image_view/image_raw', Image, callback)


  rospy.spin()

準備模型文件

wget https://paddle-inference-dist.bj.bcebos.com/Paddle-Inference-Demo/yolov3_r50vd_dcn_270e_coco.tgz &&tar xzf yolov3_r50vd_dcn_270e_coco.tgz

4.在ROS中運行Paddle Inference節點

（1）編譯

cd ../../..
catkin_make -DPYTHON_EXECUTABLE=/usr/bin/python3

（2）將工作空間添加進環境變量

sudo vim ~/.bashrc
source /home/nano/workspace/paddle_ros_ws/devel/setup.bash

（3）運行節點

roscore
rosrun py3_camera camera.py
rosrun py3_infer pp_infer.py

rosrun rqt_iamge_view rqt_iamge_view

（4）運行成功示意圖