2021-09-29 使用安卓手机Camera和IMU信息运行ORBSLAM3

目的

安卓手机具备camera imu gps等SLAM技术所需要的传感器，且安卓手机很普及,如果能使用安卓设备作为ros的sensor，通过安卓设备节点传输到计算机，进行实时定位与建图分析，那么这项技术将变得很易用。

以下介绍并不是事无巨细的介绍，要求阅读者有强大的动手能力，资料查阅能力，对linux系统，opencv，slam技术，ROS，android studio应用开发，python，C++都有一定程度的了解。

是我个人在实现过程中解决问题途径的记录，很多东西是我在网上没有找到的，希望能给大家带来帮助。

安装ros系统

ubuntu18系统下应该安装melodic

sudo sh -c '. /etc/lsb-release && echo "deb http://mirrors.ustc.edu.cn/ros/ubuntu/ $DISTRIB_CODENAME main" > /etc/apt/sources.list.d/ros-latest.list'

sudo apt update

sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys F42ED6FBAB17C654


sudo apt-get install ros-melodic-desktop-full

sudo apt install python-rosdep2

ros系统常见用法介绍

启动ros内核

roscore

查看topic列表

rostopic list

显示某个topic

rostopic echo /camera_raw

显示某个topic的header

rostopic echo /camera_raw/header

将compressed格式转为raw格式

rosrun image_transport republish compressed in:=/Mate/camera/ raw out:=/Mate/camera_raw/

限制topic频率

rosrun topic_tools throttle messages /Mate/camera_raw 4.0 /left

录制rosbag包

rosbag record -O bagpath.bag /camera_raw /imu

播放与加速播放rosbag

rosbag play -r 100 bagpath.bag

rosbag play bagpath.bag

安装rviz

----------------------

启动rviz

rviz

安卓手机充当ros sensor

安卓ROS SENSOR开源库链接

----------------------

这里注意修改相机分辨率到合适的大小，本文选择640x480

需要使用android studio同时挂好梯子，将android studio中的代理设置好

Kalibr与imu-tools对sensor进行imu和相机的联合

Kalib的个人发现难以直接使用

需要git clone进行编译

参考链接如下

编译过程中坑很多，如果遇到如找不到lblas等报错，可以考虑apt remove相关库后重新安装

过程中需要apt安装大量库

需要用到catkin(ros专用编译系统)

需要建立catkin空间，并在空间目录下新建src文件夹用于存放源码，切换到空间目录下可以执行指令进行编译，编译完成后注意添加系统变量。

生成标定版的python程序

# 如果在安装pip2包时出现了报错，可以尝试在前面加上PIP_NO_CACHE_DIR=off
PIP_NO_CACHE_DIR=off pip2 install pyX==0.12
python gen_qipan.py out.pdf --type apriltag --nx 3 --ny 4 --tsize 0.04 --tspace 0.2

我重写了一版本Python3的

----------------------

imu-tools编译与使用

也是使用catkin进行编译

中间坑挺多的

imu-tools对IMU进行标定

录制一个将SENSOR静止放置一段时间，推荐2小时的IMU数据bag

随后标定得到重力与加速度的噪声，随机走动等参数

相机标定

这里可以录制图片bag包，使用Kalibr中的程序进行标定，输出相机标定结果文件

为了提高标定效果，这里推荐如下进行操作

并使用Kalib中的提取程序提取得到图片

并使用matlab进行标定，得到的数据如何转化为Kalib和SLAM需要的格式可以参考我的另一篇文章

得到相机标定数据后，替换到相机标定结果文件中

相机与IMU联合标定

从安卓手机中直接抓取到的bag包有以下问题存在

1. IMU信息无法被Kalib直接读取
2. IMU和CAMERA的时间戳可能不在一个范围
3. IMU和CAMERA的时间戳可能不同步

这里分别有两个脚本解决以上问题

对于IMU信息无法直接读取的，使用以下脚本，在python2下运行

python2 fix_bag_msg_def.py -l inputbagpath fixedbagpath

#fix_bag_msg_def.py
import argparse
import os
import sys

try:
    import roslib.message
except:
    sys.stderr.write("Could not import 'roslib', make sure it is installed, "
        "and make sure you have sourced the ROS environment setup file if "
        "necessary.\n\n")
    sys.exit(1)

try:
    import rosbag
except:
    sys.stderr.write("Could not import 'rosbag', make sure it is installed, "
        "and make sure you have sourced the ROS environment setup file if "
        "necessary.\n\n")
    sys.exit(1)


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument('-v', '--verbose', action='store_true', help='Be verbose')
    parser.add_argument('-l', '--use-local-defs', dest='use_local', action='store_true', help='Use message defs from local system (as opposed to reading them from the provided mappings)')
    parser.add_argument('-c', '--callerid', type=str, help='Callerid (ie: publisher)')
    parser.add_argument('-m', '--map', dest='mappings', type=str, nargs=1, action='append', help='Mapping topic type -> good msg def (multiple allowed)', default=[])
    parser.add_argument('inbag', help='Input bagfile')
    parser.add_argument('outbag', help='Output bagfile')
    args = parser.parse_args()

    if not os.path.isfile(args.inbag):
        sys.stderr.write('Cannot locate input bag file [%s]\n' % args.inbag)
        sys.exit(os.EX_USAGE)

    if os.path.realpath(args.inbag) == os.path.realpath(args.outbag):
        sys.stderr.write('Cannot use same file as input and output [%s]\n' % args.inbag)
        sys.exit(os.EX_USAGE)

    if len(args.mappings) > 0 and args.use_local:
        sys.stderr.write("Cannot use both mappings and local defs.\n")
        sys.exit(os.EX_USAGE)


    # TODO: make this nicer. Figure out the complete msg text without relying on external files
    msg_def_maps = {}
    if len(args.mappings) > 0:
        print ("Mappings provided:")
        for mapping in args.mappings:
            map_msg, map_file = mapping[0].split(':')
            print ("  {:40s}: {}".format(map_msg, map_file))

            # 'geometry_msgs/PoseStamped:geometry_msgs_pose_stamped_good.txt'
            with open(map_file, 'r') as f:
                new_def = f.read()
                # skip first line, it contains something like '[geometry_msgs/PoseStamped]:'
                msg_def_maps[map_msg] = new_def.split('\n', 1)[1]
                #print (msg_def_maps[map_msg])

    else:
        if not args.use_local:
            print ("No mappings provided and not allowed to use local msg defs. "
                   "That is ok, but this won't fix anything like this.")

    print ("")


    # open bag to fix
    bag = rosbag.Bag(args.inbag)

    # filter for all connections that pass the filter expression
    # if no 'callerid' specified, returns all connections
    conxs = bag._get_connections(connection_filter=
        lambda topic, datatype, md5sum, msg_def, header:
            header['callerid'] == args.callerid if args.callerid else True)

    conxs = list(conxs)

    if not conxs:
        print ("No topics found for callerid '{}'. Make sure it is correct.\n".format(args.callerid))
        sys.exit(1)


    def_replaced = []
    def_not_found = []
    def_not_replaced = []

    # loop over connections, find out which msg type they use and replace
    # msg defs if needed. Note: this is a rather primitive way to approach
    # this and absolutely not guaranteed to work.
    # It does work for me though ..
    for conx in conxs:
        # see if we have a mapping for that
        msg_type = conx.datatype
        if not msg_type in msg_def_maps:
            if not args.use_local:
                def_not_found.append((conx.topic, msg_type))
                continue

            # don't have mapping, but are allowed to use local msg def: retrieve
            # TODO: properly deal with get_message_class failing
            sys_class = roslib.message.get_message_class(msg_type)
            if sys_class is None:
                raise ValueError("Message class '" + msg_type + "' not found.")
            msg_def_maps[conx.datatype] = sys_class._full_text

        # here, we either already had a mapping or one was just created
        full_msg_text = msg_def_maps[msg_type]

        # don't touch anything if not needed (note: primitive check)
        if conx.header['message_definition'] == full_msg_text:
            def_not_replaced.append((conx.topic, msg_type))
            continue

        # here we really should replace the msg def, so do it
        conx.header['message_definition'] = full_msg_text
        conx.msg_def = full_msg_text

        def_replaced.append((conx.topic, msg_type))


    # print stats
    if def_replaced and args.verbose:
        print ("Replaced {} message definition(s):".format(len(def_replaced)))
        for topic, mdef in def_replaced:
            print ("  {:40s} : {}".format(mdef, topic))
        print ("")

    if def_not_replaced and args.verbose:
        print ("Skipped {} message definition(s) (already ok):".format(len(def_not_replaced)))
        for topic, mdef in def_not_replaced:
            print ("  {:40s} : {}".format(mdef, topic))
        print ("")

    if def_not_found and args.verbose:
        print ("Could not find {} message definition(s):".format(len(def_not_found)))
        for topic, mdef in def_not_found:
            print ("  {:40s} : {}".format(mdef, topic))
        print ("")



    print ("Writing out fixed bag ..")

    # write result to new bag
    # TODO: can this be done more efficiently? We only changed the connection infos.
    with rosbag.Bag(args.outbag, 'w') as outbag:
        # shamelessly copied from Rosbag itself
        meter = rosbag.rosbag_main.ProgressMeter(outbag.filename, bag._uncompressed_size)
        total_bytes = 0
        for topic, raw_msg, t in bag.read_messages(raw=True):
            msg_type, serialized_bytes, md5sum, pos, pytype = raw_msg

            outbag.write(topic, raw_msg, t, raw=True)

            total_bytes += len(serialized_bytes)
            meter.step(total_bytes)

        meter.finish()

    print ("\ndone")
    print ("\nThe new bag probably needs to be re-indexed. Use 'rosbag reindex {}' for that.\n".format(outbag.filename))


if __name__ == '__main__':
    main()

对于2，3两个问题，需要使用以下脚本来解决

# roasbag_timeduiqi.py
import rospy
import rosbag
import sys

if sys.getdefaultencoding() != 'utf-8':
    reload(sys)
    sys.setdefaultencoding('utf-8')
bag_name = 'camfix.bag'
out_bag_name = '/home/leo/camduiqi.bag'
dst_dir = '/home/leo/'

with rosbag.Bag(out_bag_name, 'w') as outbag:
    stamp = None
    for topic, msg, t in rosbag.Bag(dst_dir+bag_name).read_messages():
        if topic == '/Mate/imu':
            imu_flag=True
            t_old = t
            old_stamp=msg.header.stamp

        # elif topic == '/cam0/image_raw':
        #     outbag.write(topic, msg, msg.stamp)
        #     continue
        # print msg.header
        print topic, msg.header.stamp, t
        if imu_flag and topic!="/Mate/imu":

            msg.header.stamp=old_stamp
            outbag.write(topic, msg, t_old)
            # imu_flag=False
        else:
            outbag.write(topic, msg, t)

print("finished")

这里简单介绍下原理

我的手机华为Mate30,IMU的频率为100，相机频率30

需要将相机的时间戳对齐到IMU中，这里将遍历ROSBAG中的所有信息

对于CAMERA信息，我们将其时间戳跟改为上一个IMU的时间戳，这样实现了CAMERA和IMU的时间戳对齐

ORBSLAM3的编译

这里需要编译安装opencv，等一系列库，同时也需要更改一些代码

在ROS相关的源码中，需要将topic根据实际情况进行改变

同时，可以用手机录制一个具有camera和IMU的bag，并用单目和单目+IMU进行测试

注意，这个录制的bag，也需要经过两个脚本以处理时间戳问题

参考链接

(29条消息) Kalibr 制作采集的图片和IMU数据生成 .bag 文件 （坑最全，解决方法最简单，最详细哈哈）_努力努力努力-CSDN博客https://blog.csdn.net/hltt3838/article/details/116064390