【ROS】OpenCV+ROS 实现人脸识别（Ubantu16.04）

#!/usr/bin/env python

# -*- coding: utf-8 -*-

import rospy

import cv2

import numpy as np

from sensor_msgs.msg import Image, RegionOfInterest

from cv_bridge import CvBridge, CvBridgeError

class faceDetector:

    def __init__(self):

        rospy.on_shutdown(self.cleanup);

        # 创建cv_bridge

        self.bridge = CvBridge()

        self.image_pub = rospy.Publisher("cv_bridge_image", Image, queue_size=1)

        # 获取haar特征的级联表的XML文件，文件路径在launch文件中传入

        cascade_1 = rospy.get_param("~cascade_1", "")

        cascade_2 = rospy.get_param("~cascade_2", "")

        # 使用级联表初始化haar特征检测器

        self.cascade_1 = cv2.CascadeClassifier(cascade_1)

        self.cascade_2 = cv2.CascadeClassifier(cascade_2)

        # 设置级联表的参数，优化人脸识别，可以在launch文件中重新配置

        self.haar_scaleFactor  = rospy.get_param("~haar_scaleFactor", 1.2)

        self.haar_minNeighbors = rospy.get_param("~haar_minNeighbors", 2)

        self.haar_minSize      = rospy.get_param("~haar_minSize", 40)

        self.haar_maxSize      = rospy.get_param("~haar_maxSize", 60)

        self.color = (50, 255, 50)

        # 初始化订阅rgb格式图像数据的订阅者，此处图像topic的话题名可以在launch文件中重映射

        self.image_sub = rospy.Subscriber("input_rgb_image", Image, self.image_callback, queue_size=1)

    def image_callback(self, data):

        # 使用cv_bridge将ROS的图像数据转换成OpenCV的图像格式

        try:

            cv_image = self.bridge.imgmsg_to_cv2(data, "bgr8")     

            frame = np.array(cv_image, dtype=np.uint8)

        except CvBridgeError, e:

            print e

        # 创建灰度图像

        grey_image = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)

        # 创建平衡直方图，减少光线影响

        grey_image = cv2.equalizeHist(grey_image)

        # 尝试检测人脸

        faces_result = self.detect_face(grey_image)

        # 在opencv的窗口中框出所有人脸区域

        if len(faces_result)>0:

            for face in faces_result:

                x, y, w, h = face

                cv2.rectangle(cv_image, (x, y), (x+w, y+h), self.color, 2)

        # 将识别后的图像转换成ROS消息并发布

        self.image_pub.publish(self.bridge.cv2_to_imgmsg(cv_image, "bgr8"))

    def detect_face(self, input_image):

        # 首先匹配正面人脸的模型

        if self.cascade_1:

            faces = self.cascade_1.detectMultiScale(input_image,

                    self.haar_scaleFactor,

                    self.haar_minNeighbors,

                    cv2.CASCADE_SCALE_IMAGE,

                    (self.haar_minSize, self.haar_maxSize))

        # 如果正面人脸匹配失败，那么就尝试匹配侧面人脸的模型

        if len(faces) == 跟单网gendan5.com0 and self.cascade_2:

            faces = self.cascade_2.detectMultiScale(input_image,

                    self.haar_scaleFactor,

                    self.haar_minNeighbors,

                    cv2.CASCADE_SCALE_IMAGE,

                    (self.haar_minSize, self.haar_maxSize))

        return faces

    def cleanup(self):

        print "Shutting down vision node."

        cv2.destroyAllWindows()

if __name__ == '__main__':

    try:

        # 初始化ros节点

        rospy.init_node("face_detector")

        faceDetector()

        rospy.loginfo("Face detector is started..")

        rospy.loginfo("Please subscribe the ROS image.")

        rospy.spin()

    except KeyboardInterrupt:

        print "Shutting down face detector node."

        cv2.destroyAllWindows()