solbach/fallen-person-detector

Vision-Based Fallen Person Detection for the Elderly

fallen-person-detector's Issues

Not detecting fallen person

When I run the fallen_person_detector node it doesn't detect a fallen person.
For the following cloudGround point cloud(fallen person clearly visible),

I get the following CoG and UbC,

[ INFO] [1519144336.501612941]: CoG distance to ground: 2.911256m
[ INFO] [1519144336.501796666]: UbC distance to ground: 2.933364m

Also, to avoid getting 999999.0 CoG and UbC due to nan in body parts point cloud, I have removed them before the centroid is calculated.
I've commented out the piece of code which calculates Area and Volume, since I was getting a segmentation fault, and also these values aren't used anywhere, how can they be used for debugging purposes.

I also visualized extracted ground from cloudGround which wasn't as I expected(not representing the ground) and so I wanted to ask if there are some parameters that needs to be changed for the morphological filter, or does it work for specific environments.

Using Ubuntu 16.04, with OpenCV 3.2, CUDA 8.0 and Pepper Robot.
Any suggestions?
Thanks

Trying to extract 3D poses

I am trying to get 3d poses out by publishing the point cloud including 3D body poses to a topic. But I get xW, yW, zW values all as 0.
Also adding
ROS_INFO( "BODYPARTS\n%s,Confidence,x,y\t%lf,%lf,%lf\n%lf\n",bodyparts[i].name , bodyparts[i].confidence , bodyparts[i].x ,bodyparts[i].y , depth.at<float>(bodyparts[i].y, bodyparts[i].x));
shows that though for all is bodyparts[i].confidence has some value but all bodyparts[i].x,bodyparts[i].y are 0.

Running steps

After running openpose_ros and fallen_person_detector, I get the following output,

And nothing happens, any suggestions on running the model on custom images? Thanks