CSCE 452/752: Project 3

Overview and purpose

This assignment asks you to create software that processes data from a lidar sensor to detect and track the movement of people around a robot. The purpose is to give some experience designing and implementing algorithms that cope with the complexity of real sensor data, along with exposure to some additional ROS features.

Reference material

In addition to the concepts you used in Projects 1 and 2 (nodes, topics, publishers, subscribers, services, parameters, etc.), you'll need to use launch files and visualization markers. You will likely want to consult the tutorials and reference documentation on those subjects.

Your task

Here is a collection bag files recorded by a robot equipped with a laser range finder (i.e. lidar):

Project 3 Bag Files

When each of these bags was recorded, the robot and the nearby obstacles remained motionless. However, in each case, zero or more people walked past the robot.

Your job is create a system to detect and track the unique people that the robot sees.

Several times per second, your system should publish a message of type

visualization_msgs/MarkerArray

on the topic

/person_markers

. This message should contain one marker for each person that has passed in front of the sensor, including people currently in view of the sensor and people that passed in front of the sensor in the past.

The marker type should be
LINE_STRIP
, and the other fields of the marker object should be filled appropriately in to enable
rviz
to show the history of that person's movements.
Each mark should start from the place where the person first appeared.
For people currently in the scene, the marker should show their movements up until the present time. The
color
for these markers must have an
r
value less than 0.2.
For people not currently in the scene, the marker should show their movements up until the last location where they were observed. The
color
for these markers must have an
r
value greater than 0.8.
Apart from using the
r
values to indicate which people are currently in the scene, the markers may use any colors you like.
To keep track of which person is which, the
id
field of each marker should uniquely and persistently identify each person. The specific id numbers you use are not important, as long as the
id
remains consistent for each person and distinct people are assigned distinct
id
s.

An example with three markers appears below.

Your system must consist of at least two distinct nodes. One node should subscribe to

/scan

and a different node or nodes should publish the markers. You may use more than two nodes if you wish. You should design your system carefully to make a meaningful subdivision of the work between the nodes in your system. Your nodes should communicate by publishing messages on a topic or topics of your own design. (The purpose of this constraint is to practice the ROS philosophy of designing nodes that each do a single, simple job well, and then connecting those parts together to form a complete system.)

Finally, because your system will consist of several distinct components that must be started in concert, you should create a launch file that starts the system. This launch file should:

Have an argument called
bag_in
. The argument will be set on the
roslaunch
command line to the name of a bag file (one of the provided bags) to play.
Have an argument called
bag_out
. The argument will be set on the
roslaunch
command line to the name of the bag file to record all published messages to.
Start
ros2 bag play
to play the provided sensor data,
ros2 bag record
to capture all topics, including the output for your system, along with all of the nodes needed to execute your system.
Terminate all of the other nodes when
rosbag play
completes.

Your solution will be evaluated primarily on the accuracy of its results. The quality of the system design and the code will also play roles in the grading.

What to submit

Three required parts:

A PDF report including:
1. A concise description of how the system works. This should be a couple of paragraphs, describing your overall approach at a conceptual level. Some questions to consider: What choices did you make in designing the program, and why? What nodes does your system contain, and what role does each one play? What specific parameters are part of your method, and how did you select values for those parameters?
2. A short answer to these questions: Did the results meet your expectations? Why or why not?
The source code for your programs. This will span at least two source code files. Be sure to consult the source code style guide on the course Canvas page.
Rosbag recordings of all topics throughout the entire execution of your program for each of the provided input bags.
It is essential for these bags to contain both the original sensor data on
/scan
as well as the outputs on
/person_markers
because this information will play a primary role in evaluating the correctness of your solution. We cannot award credit for correctly detecting and tracking people unless you submit these outputs correctly.
The directory of each bag file must be named with your TAMU netid. followed by a dash, followed by the number of the input bag file. For example, your instructor's submission would include bag directories called
jokane-1
,
jokane-2
, and so on. Incorrectly named bags cannot be graded.
Each rosbag recording is a directory which contains both a
db3
or
mcap
file and a yaml file of metadata. You may rename this directory, but you should not modify the names or contents of the files after recording, or you risk corrupting the rosbag. Corrupted rosbags cannot be graded. You should include the entire directory in your submission.

One optional part:

A short video demonstrating your system in some way that may be helpful in grading.