Angus's profile - activity

In C++, variable-length ROS message types are implemented as std::vector (source). So you should be able to do

for (int i = 0; i < msg->ranges.size(); i++)
  float r = msg->ranges[i];

or

for (std::vector<float>::iterator it = msg->ranges.begin(); it != ms->ranges.end(); ++it)
  float r = *it;

Disclaimer: I haven't actually tried running this exact code.

Hi Phelipe!

To count the number of detections from the HOG detector use:

int num_detected_persons = detected_person.size();

To count the number of detections from the HAAR detector use:

int num_detected_faces = detected_faces.size();

An alternative for the kinect which I would recommended trying out is the ground-based RGBD detection in the PCL by Matteo Munaro

How are you filtering out the 0.0m measurements? Maybe you or gmapping is setting those points to scan.range_max + 1 as in the laser scan range filter? Can you see any other conspicuous ranges, such as NaNs, infs, or -infs in the laser scan?

It can detect both. Give it a shot but don't expect very robust performance from it unless you combine it with a tracking-by-detection method (e.g., a Kalman filter with outlier rejection), geometric constraints (e.g., filtering based on size) or background subtraction.

Try:

sudo apt-get install ros-indigo-bfl

Yup, see Table 5-1 on page 11 of the manual. Its max scanning angle is limited to 100 degrees when using an angular resolution of 0.25 degrees.

What happens when you try the commands here?

Awesome! I'll just reply with an answer so we can mark as solved.

As per the comments, the demo seems to work on Inidigo with these commands:

$ cd [indigo catkin workspace directory]
$ git clone https://github.com/ccny-ros-pkg/scan_tools
$ sudo apt-get install libgsl0-dev
$ catkin_make
$ roslaunch laser_scan_matcher demo.launch

I just got the demo working on my machine with:

• $ cd [indigo catkin workspace directory]
• $ git clone https://github.com/ccny-ros-pkg/scan_...
• $ sudo apt-get install libgsl0-dev
• $ catkin_make
• $ roslaunch laser_scan_matcher demo.launch

Is this what you did?

I don't believe it supports that resolution/scan angle combination. See page 11: http://sicktoolbox.sourceforge.net/docs/sick-lms-technical-description.pdf

I fixed my "undefined reference to tf::Transformer..." by including "tf" in the "find_package" part of the CMakeLists.

It appears that in the transformLaserScanToPointCloud function a lookupTransform is called at time:

scan_in.header.stamp + ros::Duration ().fromSec (scan_in.ranges.size () * scan_in.time_increment)

My quick-fix solution was to call waitForTransform and include that added time before calling transformLaserScanToPointCloud.

Source: https://github.com/ros-perception/las...

And I think this is the paper's ROS webpage: http://wiki.ros.org/cyphy_people_mapping

"Although early versions of the OpenNI skeleton tracker required users to surrender to the kinect in order for tracking to begin, recent versions allow the saving and loading of user calibration files."

Slight lead that may or may not help you out: in the paper "STALKERBOT: Learning to Navigate Dynamic Human Environments by Following People" they say:

You might want to check out HOG detector for OpenCV which tries to detect standing pedestrians, although you're going to get a lot of false positives and missed detections from it. Also, it seems they only have documentation on the website for the GPU implementation (i.e. using CUDA) and you'll have to look elsewhere for examples of the CPU implementation.

Another possibility to check out is the OpenCV face detector. Some ROS packages that use this are face_detector (uses stereo images though) and pi_face_tracker (for Groovy though).

In general, detecting people from a moving robot with a monocular camera is still a very difficult and open problem. If your camera is stationary then the problem can be simplified through background subtraction but I don't have any references for that.

For help with using OpenCV in ROS, checkout the vision_opencv package.

Edit: I uploaded my implementation of both of these in a ROS package on Github: https://github.com/angusleigh/hog_haa...

Your main "while" loop will start running immediately after launching the node, but the values for "r_laser" will not be set until the callback is made, so when you try to access the first element, you get a segfault. Maybe try changing the code in your main loop to

if (!r_laser.ranges.empty())
  ROS_INFO("point_of_ranges=[%f] \n", r_laser.ranges[0])

This is common bug with Rviz 1.9.34 in Groovy for non-Nvidia video cards. If you do a sudo apt-get update does it update to Rviz 1.9.35 for you? I would recommend that, switching to Hydro, or compiling one of the other versions of Rviz from source (maybe 1.9.33 or 1.9.35).

Which ROS version? Which Rviz version (you can check by $rosrun rviz rviz)?

Looking at the outputs from the leg_detector package, I'm wondering if it could benefit from a more diverse set of training examples. Notably, the detection confidence falls quickly as people move away from the laser and it is fooled easily by round objects like chair legs.

Does anyone know if the original training data is available anywhere? Maybe also the labelling/annotation method that was used?

Relevant links:

• here's an unofficial repo of leg_detector/leg_detector2.
• this is David Lu's fork of leg_detector
• this is the SRS leg_detector fork

According to the gazebosim wiki, version 1.9.x of gazebo should be installed when ros hydro is installed but for reasons unknown I'm getting gazebo-current (i.e. version 2.x.y) installed. Any idea why?

I'm entering:

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

and the output is:

The following NEW packages will be installed:
gazebo-current ...

I've also had gazebo-current install earlier but have since removed it.

Hello,

Does anyone know if there's any way to insert breakpoints and step through a Gazebo plugin? This would be very useful to help with debugging logical errors. I've tried inserting breakpoints in the eclipse editor, compiling and running the launch file but it doesn't seem to work. As such, I've just resorted to printf([var])'s for now.

Thanks very much!

Angus