• 本實驗采用ROS和OpenCV實作功能，實驗平臺采用Parrot的Bebop2無人機
• ROS部分的學習可以參考我的專欄：ROS學習記錄
• 實驗平臺的操作方式見：ROS控制Parrot Bebop2無人機

1. 物體跟蹤

1.1 實作思路

呼叫無人機的影像：

cv_image = self.bridge.imgmsg_to_cv2(data, “bgr8”)

之后同OpenCV實作機器人對物體進行移動跟隨一樣，獲取所要跟蹤的物體

節點的發布和接收見：ROS學習: Topic通訊

1.2 代碼示例

import rospy
import cv2 as cv
from geometry_msgs.msg import Twist
from cv_bridge import CvBridge, CvBridgeError
from sensor_msgs.msg import Image


class image_converter:
    def __init__(self):
        self.cmd_pub = rospy.Publisher("/bebop/cmd_vel", Twist, queue_size=1)			# 發布運動控制資訊
        self.bridge = CvBridge()
        self.image_sub = rospy.Subscriber("/bebop/image_raw", Image, self.callback)			# 訂閱攝像頭資訊

    def callback(self, data):
        try:
            cv_image = self.bridge.imgmsg_to_cv2(data, "bgr8")			# 獲取訂閱的攝像頭影像
        except CvBridgeError as e:
            print e

       # 對影像進行處理
        kernel = cv.getStructuringElement(cv.MORPH_RECT, (3, 3))			 # 定義結構元素
        height, width = cv_image.shape[0:2]
        screen_center = width / 2
        screen_center_h = height / 2
        
        offset = 50
        offset_h = 30
        lower_b = (75, 43, 46)
        upper_b = (110, 255, 255)
        
        hsv_frame = cv.cvtColor(cv_image, cv.COLOR_BGR2HSV)				# 轉成HSV顏色空間
        mask = cv.inRange(hsv_frame, lower_b, upper_b)								
        mask2 = cv.morphologyEx(mask, cv.MORPH_OPEN, kernel)			# 開運算去噪
        mask3 = cv.morphologyEx(mask2, cv.MORPH_CLOSE, kernel)		# 閉運算去噪
        cv.imshow("mask", mask3)
        
        # 找出面積最大的區域
        _, contours, _ = cv.findContours(mask3, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
        maxArea = 0
        maxIndex = 0
        for i, c in enumerate(contours):
            area = cv.contourArea(c)
            if area > maxArea:
                maxArea = area
                maxIndex = i
        # 繪制輪廓
        cv.drawContours(cv_image, contours, maxIndex, (255, 255, 0), 2)			

        # 獲取外切矩形
        x, y, w, h = cv.boundingRect(contours[maxIndex])
        cv.rectangle(cv_image, (x, y), (x + w, y + h), (255, 0, 0), 2)
        # 獲取中心像素點
        center_x = int(x + w / 2)
        center_y = int(y + h / 2)
        cv.circle(cv_image, (center_x, center_y), 5, (0, 0, 255), -1)
		# 顯示影像
        cv.imshow("Image", cv_image)

        # 運動控制
        twist = Twist()
        # 左右轉向和移動
        if center_x < screen_center - offset:
            twist.linear.x = 0.0
            twist.linear.y = 0.2
            twist.angular.z = 0.2
            print "turn left"
        elif screen_center - offset <= center_x <= screen_center + offset:
            twist.linear.x = 0.0
            twist.linear.y = 0.0
            twist.angular.z = 0
            print "keep"
        elif center_x > screen_center + offset:
            twist.linear.x = 0.0
            twist.linear.y = -0.2
            twist.angular.z = -0.2
            print "turn right"
        else:
            twist.linear.x = 0
            twist.angular.z = 0
            print "stop"
		# 上下移動
        if center_y < screen_center_h - offset_h:
            twist.linear.z = 0.2
            print "up up up"
        elif screen_center_h - offset_h <= center_y <= screen_center_h + offset_h:
            twist.linear.z = 0
            print "keep"
        elif center_y > screen_center_h + offset_h:
            twist.linear.z = -0.2
            print "down down down"
        else:
            twist.linear.z = 0
            print "stop"
        cv.waitKey(3)
		# 發布運動指令
        try:
            self.cmd_pub.publish(twist)
        except CvBridgeError as e:
            print e


if __name__ == '__main__':
    try:
        rospy.init_node("cv_bridge_test")
        rospy.loginfo("Starting cv_bridge_test node")
        image_converter()
        rospy.spin()

    except KeyboardInterrupt:
        print "Shutting down cv_bridge_test node."
        cv.destroyAllWindows()

效果圖

2. 自主尋線

將上節的物體識別改為所尋線，運動控制左右移動/轉向，剩下就是調參的事情了