【从kitti开始自动驾驶】--7.2 实现3d侦测框(移植至ROS,显示于RVIZ)

本节将采用在jupyter notebook上的方式执行对单张图片上实现对所有object的3D打标,以及移植到ROS上,通过RVIZ可视化出来

tracking资料存放在/kitti_folder/tracking中
jupyter工程存放在/test3_autodrive_ws/src/jupyter_prj下的2D_tracking_label.ipynb
数据文件存放在/home/qinsir/kitti_folder/tracking/data_tracking_label_2/training/label_02
下,可以看到是txt文件
矫正程序在test3_autodrive_ws/src/demo1_kitti_pub_photo/scripts/calibration.py中

上一节已经实现在jupyter-notebook平台上的稳定运行,本节将其移植至ROS中即可

1 kitti_all中的改动

1.1 准备工作

• 导入之前写好的cali的包
• 引进根据坐标定位矩形框八个顶点的坐标的函数

from calibration import *
def compute_3d_box_cam2(h, w, l, x, y, z, yaw)

---

1.2 发布者建立

box3d_pub = rospy.Publisher("kitti_3d", MarkerArray, queue_size=10)

---

1.3 矫正的instance读取

calib = Calibration('/home/qinsir/kitti_folder/2011_09_26', from_video=True)

---

1.4 所需数据提取转为数组

boxes_3d = np.array(df_tracking[df_tracking.frame==frame][['height', 'width', 'length', 'pos_x', 'pos_y', 'pos_z', 'rot_y']])

---

1.5 为每个object都画框

• 每个object迭代一次
• 第一个数组用来存放每一次经过转换的8个点的坐标
• 一个物体是8X3, 每个物体都存在corners_3d_velos中

        corners_3d_velos = []

        for box_3d in boxes_3d:
            
            corners_3d_cam2 = compute_3d_box_cam2(*box_3d)        #用*表展开为7个参数
            #经过矫正到velodyne的函数,入口参数为从cam到velodyne
            #需要8X3,转置一下
            corners_3d_velo = calib.project_ref_to_velo(corners_3d_cam2.T)
            corners_3d_velos += [corners_3d_velo]           #逐个存入列表

---

1.6 执行发布函数

• 没啥说的,因为要在publish_utils.py中定义
• 最后一个参数是为了为不同obj指定不同颜色

publish_3dbox(box3d_pub, corners_3d_velos, types)   #发布三维

2 publish_utils中的改动

2.1 变量定义

• 定义生命周期,和图片的生存周期,一秒十帧一样
• 定义每两个之间需要划线的点,用X代表前方

LIFETIME = 0.1   #我没用上

LINES = [[0,1], [1,2], [2,3], [3, 0]]   #底面
LINES += [[4,5], [5,6], [6,7], [7,4]]   #顶面
LINES += [[4,0], [5,1], [6,2], [7,3]]   #中间的四条线
LINES += [[4,1], [5,0]]                 #前边用"X"标志

---

2.2 为不同object加入不同颜色

• 因为这里的颜色范围是0-1,所以进行归一化

b, g, r = DETECTION_COLOR_DICT[types[i]]

        marker.color.r = r / 255.0
        marker.color.g = g / 255.0
        marker.color.b = b / 255.0           #这条线的颜色

2.3 maker_array操作

• 因为一次发布多条线,所以采用marker_array同时发送的方法
• 常规的,建立marker_array,marker,给marker填充数据再增加到marker_array中
• 操作较为常规,就不在此展现代码了

2.4 大遍历每一个object

• i是序号
• corners_3d_velo是该物体对应的8个点

 for i, corners_3d_velo in enumerate(corners_3d_velos):

2.5 小遍历每两个点

• 在我们定义的需要连接的点对中,进行划线
• 需要连线的一对点/一条线收录进同一个marker中,在append到maker_array中

for l in LINES:
            p1 = corners_3d_velo[l[0]]
            marker.points.append(Point(p1[0], p1[1], p1[2]))
            p2 = corners_3d_velo[l[1]]
            marker.points.append(Point(p2[0], p2[1], p2[2]))

2.6 发布

box3d_pub.publish(marker_array)

3.效果查看

4. 源码

4.1 publish_utils.py代码

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

import rospy
from std_msgs.msg import Header
from sensor_msgs.msg import Image, PointCloud2, Imu, NavSatFix
from visualization_msgs.msg import Marker, MarkerArray
import sensor_msgs.point_cloud2 as pcl2
from geometry_msgs.msg import Point
from cv_bridge import CvBridge
import numpy as np
import tf
import cv2

FRAME_ID = "map"
DETECTION_COLOR_DICT = {'Car':(255, 255, 0), 'Pedestrian':(0, 226, 255), 'Cyclist':(141, 40, 255)}
#为三种车型,分别添加颜色值对应的框
LIFETIME = 0.1 
#生存周期

LINES = [[0,1], [1,2], [2,3], [3, 0]]   #底面
LINES += [[4,5], [5,6], [6,7], [7,4]]   #顶面
LINES += [[4,0], [5,1], [6,2], [7,3]]   #中间的四条线
LINES += [[4,1], [5,0]]                 #前边用"X"标志
'''
def publish_camera(cam_pub, bridge, image):
    cam_pub.publish(bridge.cv2_to_imgmsg(image, 'bgr8'))
'''
def publish_camera(cam_pub, bridge, image, boxes, types):
    #画出每一个坐标
    for typ, box in zip(types, boxes):    #for一个矩阵,它是一行一行读取
        # 左上角,右下角,像素都是整数
        top_left = int(box[0]), int(box[1])
        right_down = int(box[2]), int(box[3])
        # 画矩形
        cv2.rectangle(image, top_left, right_down, DETECTION_COLOR_DICT[typ], 2)
    cam_pub.publish(bridge.cv2_to_imgmsg(image, 'bgr8'))

def publish_point_cloud(pcl_pub, point_cloud):
    header = Header()
    header.frame_id = FRAME_ID
    header.stamp = rospy.Time.now()
    pcl_pub.publish(pcl2.create_cloud_xyz32(header, point_cloud[:, :3]))


def publish_ego_car(ego_car_pub):
    'publish left and right 45 degree FOV and ego car model mesh'
    #header部分
    marker = Marker()
    marker.header.frame_id = FRAME_ID
    marker.header.stamp = rospy.Time.now()
    # marker的id 
    marker.id = 0
    marker.action = Marker.ADD      # 加入一个marker
    marker.lifetime = rospy.Duration()  # 生存时间,()中无参数永远出现
    marker.type = Marker.LINE_STRIP     #marker 的type,有很多种,这里选择线条

    marker.color.r = 0.0
    marker.color.g = 1.0
    marker.color.b = 0.0            #这条线的颜色
    marker.color.a = 1.0            #透明度 1不透明
    marker.scale.x = 0.2            #大小,粗细

    #设定marker中的资料
    marker.points = []
    # 两条线,三个点即可
    #原点是0,0,0, 看左上角和右上角的数据要看kitti的设定,看坐标
    marker.points.append(Point(10, -10, 0))
    marker.points.append(Point(0, 0, 0))
    marker.points.append(Point(10, 10, 0))

    ego_car_pub.publish(marker) #设定完毕,发布

def publish_car_model(model):
    #header部分
    mesh_marker = Marker()
    mesh_marker.header.frame_id = FRAME_ID
    mesh_marker.header.stamp = rospy.Time.now()
    # marker的id 
    mesh_marker.id = -1
    mesh_marker.lifetime = rospy.Duration()  # 生存时间,()中无参数永远出现
    mesh_marker.type = Marker.MESH_RESOURCE     #marker 的type,有很多种,这里选择mesh
    mesh_marker.mesh_resource = "package://demo1_kitti_pub_photo/mesh/car_model/car.DAE"

    #平移量设置
    mesh_marker.pose.position.x = 0.0
    mesh_marker.pose.position.y = 0.0
    #以为0,0,0 是velodyne的坐标(车顶),这里坐标是车底,所以是负数
    mesh_marker.pose.position.z = -1.73

    #旋转量设定
    q = tf.transformations.quaternion_from_euler(np.pi/2, 0, np.pi/2)
    # 这里的参数和下载模型的预设角度有关,旋转关系,根据显示效果而调整,转成四元数q
    #x轴旋转
    mesh_marker.pose.orientation.x = q[0]
    mesh_marker.pose.orientation.y = q[1]
    mesh_marker.pose.orientation.z = q[2]
    mesh_marker.pose.orientation.w = q[3]

    #颜色设定(白色)
    mesh_marker.color.r = 1.0
    mesh_marker.color.g = 1.0
    mesh_marker.color.b = 1.0
    mesh_marker.color.a = 1.0

    #设置体积:  x,y,z方向的膨胀程度
    mesh_marker.scale.x = 0.4
    mesh_marker.scale.y = 0.4
    mesh_marker.scale.z = 0.4

    model.publish(mesh_marker) #设定完毕,发布

def publish_two_marker(kitti_two_marker):
    #建立markerarray
    markerarray = MarkerArray()
    # 绿线设定
    marker = Marker()
    marker.header.frame_id = FRAME_ID
    marker.header.stamp = rospy.Time.now()
    # marker的id 
    marker.id = 0
    marker.action = Marker.ADD      # 加入一个marker
    marker.lifetime = rospy.Duration()  # 生存时间,()中无参数永远出现
    marker.type = Marker.LINE_STRIP     #marker 的type,有很多种,这里选择线条

    marker.color.r = 0.0
    marker.color.g = 1.0
    marker.color.b = 0.0            #这条线的颜色
    marker.color.a = 1.0            #透明度 1不透明
    marker.scale.x = 0.2            #大小,粗细

    #设定marker中的资料
    marker.points = []
    # 两条线,三个点即可
    #原点是0,0,0, 看左上角和右上角的数据要看kitti的设定,看坐标
    marker.points.append(Point(10, -10, 0))
    marker.points.append(Point(0, 0, 0))
    marker.points.append(Point(10, 10, 0))
    #加入第一个
    markerarray.markers.append(marker)

    mesh_marker = Marker()
    mesh_marker.header.frame_id = FRAME_ID
    mesh_marker.header.stamp = rospy.Time.now()
    # marker的id 
    mesh_marker.id = -1
    mesh_marker.lifetime = rospy.Duration()  # 生存时间,()中无参数永远出现
    mesh_marker.type = Marker.MESH_RESOURCE     #marker 的type,有很多种,这里选择mesh
    mesh_marker.mesh_resource = "package://demo1_kitti_pub_photo/mesh/car_model/car.DAE"

    #平移量设置
    mesh_marker.pose.position.x = 0.0
    mesh_marker.pose.position.y = 0.0
    #以为0,0,0 是velodyne的坐标(车顶),这里坐标是车底,所以是负数
    mesh_marker.pose.position.z = -1.73

    #旋转量设定
    q = tf.transformations.quaternion_from_euler(np.pi/2, 0, np.pi/2)
    # 这里的参数和下载模型的预设角度有关,旋转关系,根据显示效果而调整,转成四元数q
    #x轴旋转
    mesh_marker.pose.orientation.x = q[0]
    mesh_marker.pose.orientation.y = q[1]
    mesh_marker.pose.orientation.z = q[2]
    mesh_marker.pose.orientation.w = q[3]

    #颜色设定(白色)
    mesh_marker.color.r = 1.0
    mesh_marker.color.g = 1.0
    mesh_marker.color.b = 1.0
    mesh_marker.color.a = 1.0

    #设置体积:  x,y,z方向的膨胀程度
    mesh_marker.scale.x = 0.4
    mesh_marker.scale.y = 0.4
    mesh_marker.scale.z = 0.4
    # 加入第二个:车子模型
    markerarray.markers.append(mesh_marker)

    #发布
    kitti_two_marker.publish(markerarray)

def publish_imu(imu_pub, imu_data):
    # 消息建立
    imu = Imu()

    #头填充
    imu.header.frame_id = FRAME_ID
    imu.header.stamp = rospy.Time.now()

    #旋转角度
    q = tf.transformations.quaternion_from_euler(float(imu_data.roll), float(imu_data.pitch), float(imu_data.yaw))
    # 将roll, pitch, yaw转成可被电脑识别的四元数q,并设定出去
    imu.orientation.x = q[0]
    imu.orientation.y = q[1]
    imu.orientation.z = q[2]
    imu.orientation.w = q[3]
    #线性加速度
    imu.linear_acceleration.x = imu_data.af
    imu.linear_acceleration.y = imu_data.al
    imu.linear_acceleration.z = imu_data.au 
    #角速度
    imu.angular_velocity.x = imu_data.wf
    imu.angular_velocity.y = imu_data.wl
    imu.angular_velocity.z = imu_data.wu

    #发布
    imu_pub.publish(imu)

def  publish_gps(gps_pub, imu_data):
    gps = NavSatFix()

    #头填充
    gps.header.frame_id = FRAME_ID
    gps.header.stamp = rospy.Time.now()

    #维度, 经度 和海拔
    gps.latitude = imu_data.lat
    gps.longitude = imu_data.lon
    gps.altitude = imu_data.al

    gps_pub.publish(gps)

def publish_3dbox(box3d_pub, corners_3d_velos, types):
    marker_array = MarkerArray()
    #对每一组数据/交通工具/矩形框进行迭代
    for i, corners_3d_velo in enumerate(corners_3d_velos):
        #header部分
        marker = Marker()
        marker.header.frame_id = FRAME_ID
        marker.header.stamp = rospy.Time.now()

        # marker的id 
        marker.id = i
        marker.action = Marker.ADD      # 加入一个marker
        marker.lifetime = rospy.Duration(0)  # 生存时间,()中无参数永远出现
        marker.type = Marker.LINE_STRIP     #marker 的type,有很多种,这里选择线条

        b, g, r = DETECTION_COLOR_DICT[types[i]]

        marker.color.r = r / 255.0
        marker.color.g = g / 255.0
        marker.color.b = b / 255.0           #这条线的颜色

        marker.color.a = 1.0            #透明度 1不透明

        marker.scale.x = 0.1            #大小,粗细

        #设定marker中的资料
        marker.points = []
        #对每两个点之间进行迭代,存在则连
        for l in LINES :
            p1 = corners_3d_velo[l[0]]
            marker.points.append(Point(p1[0], p1[1], p1[2]))
            p2 = corners_3d_velo[l[1]]
            marker.points.append(Point(p2[0], p2[1], p2[2]))

        marker_array.markers.append(marker)

    
    box3d_pub.publish(marker_array)

4.2 kitti_all.py

#!/usr/bin/env python3
#coding:utf-8
from data_utils import *
from publish_utils import *
from calibration import *


DATA_PATH = '/home/qinsir/kitti_folder/2011_09_26/2011_09_26_drive_0005_sync/'

# 根据长宽高XYZ和旋转角坐标定位
def compute_3d_box_cam2(h, w, l, x, y, z, yaw):
    '''
    return : 3xn in cam2 coordinate
    '''
    
    R = np.array([[np.cos(yaw), 0, np.sin(yaw)], [0, 1, 0], [-np.sin(yaw), 0, np.cos(yaw)]])
    x_corners = [l/2, l/2, -l/2, -l/2, l/2, l/2, -l/2, -l/2]
    y_corners = [0, 0, 0, 0, -h, -h, -h, -h]
    z_corners = [w/2, -w/2, -w/2, w/2, w/2, -w/2, -w/2, w/2]
    
    corners_3d_cam2 = np.dot(R, np.vstack([x_corners, y_corners, z_corners]))
    corners_3d_cam2 += np.vstack([x, y, z])
    
    return corners_3d_cam2

if __name__ == '__main__':
    frame = 0
    rospy.init_node('kitti_node', anonymous=True)   #默认节点可以重名
    cam_pub = rospy.Publisher('kitti_cam', Image, queue_size=10)
    pcl_pub = rospy.Publisher("kitti_point_cloud", PointCloud2, queue_size=10)
    #ego_pub = rospy.Publisher('kitti_ego_car', Marker, queue_size=10)
    #model_pub = rospy.Publisher("kitti_car_model", Marker, queue_size=10)
    two_marker_pub = rospy.Publisher("kitti_two_mark", MarkerArray, queue_size=10)
    imu_pub = rospy.Publisher("kitti_imu", Imu, queue_size=10)  #IMU发布者
    gps_pub = rospy.Publisher("kitti_gps", NavSatFix, queue_size=10)
    box3d_pub = rospy.Publisher("kitti_3d", MarkerArray, queue_size=10)

    bridge = CvBridge()      #opencv支持的图片和ROS可以读取的图片之间转换的桥梁

    rate = rospy.Rate(10)
    #以第0组(厢型车等归类适用的tracking数据)为数据读取
    df_tracking = read_tracking('/home/qinsir/kitti_folder/tracking/data_tracking_label_2/training/label_02/0000.txt')
    calib = Calibration('/home/qinsir/kitti_folder/2011_09_26', from_video=True)

    while not rospy.is_shutdown():
        boxes_2d = np.array(df_tracking[df_tracking.frame==frame][["bbox_left", "bbox_top", "bbox_right", "bbox_bottom"]])  #格式转换,第0帧所有的打标框
        types = np.array(df_tracking[df_tracking.frame==frame]["type"])
        boxes_3d = np.array(df_tracking[df_tracking.frame==frame][['height', 'width', 'length', 'pos_x', 'pos_y', 'pos_z', 'rot_y']])

        #保存8个雷达坐标系下点的坐标的list
        corners_3d_velos = []
        #对于图片中的每一个记录物体都进行计算
        for box_3d in boxes_3d:
            #相机里的坐标
            corners_3d_cam2 = compute_3d_box_cam2(*box_3d)        #用*表展开为7个参数
            #经过矫正到velodyne的函数,入口参数为从cam到velodyne
            #需要8X3,转置一下
            corners_3d_velo = calib.project_ref_to_velo(corners_3d_cam2.T)
            corners_3d_velos += [corners_3d_velo]           #逐个存入列表


        #使用OS,路径串接,%010d,这个字串有10个数字(比如0000000001).png
        img = read_camera(os.path.join(DATA_PATH, 'image_02/data/%010d.png'%frame)) 
        point_cloud = read_point_cloud(os.path.join(DATA_PATH, "velodyne_points/data/%010d.bin"%frame))
        imu_data = read_imu(os.path.join(DATA_PATH, "oxts/data/%010d.txt"%frame))

        publish_camera(cam_pub, bridge, img, boxes_2d, types)
        #publish_camera(cam_pub, bridge, img)
        publish_point_cloud(pcl_pub, point_cloud)
        publish_3dbox(box3d_pub, corners_3d_velos, types)   #发布三维
        #publish_ego_car(ego_pub)
        #publish_car_model(model_pub)
        publish_two_marker(two_marker_pub)
        publish_imu(imu_pub, imu_data)
        publish_gps(gps_pub, imu_data)

        rospy.loginfo('new file publish')
        rate.sleep()
        frame += 1
        frame %= 154