Does anyone know a good algorithm for calculating the closest points/distance between two line segments? I use some pretty general code:
http://geomalgorithms.com/a07-_distance.html
which seems widely used and advertized on the web and works in most cases but seems to often fail horribly when line segments are nearly parallel. I've been messing with the SMALL_NUM value for division overflow to no avail. The calculated distance can still vary widely when nearly parallel.
I managed to isolate a specific incident where this happens in my code. The distance between segments P1P2 and Q1Q2 changed abruptly in one timestep from 1.05 mm to 0.90 mm (yarn radius = 1 mm), causing abrupt compression spikes. In reality the distance in the original timestep is definitely also around 0.90 mm but is not calculated as such. I find that the values of s and t (s=0 for P1, 1 for P2, t=0 for Q1, 1 for Q2) for the closest points are originally 0 and 0 (as well as in the previous time steps) and then change abruptly to 0 and around 0.29 in the new time step. What it should be, I still need to check out.
P1 = (0.012711 ,-0.000688 ,-0.001097);
P2 = (0.012895 ,-0.000686 ,-0.001133);
Q1 = (0.012676 ,-0.000689 ,-0.000999);
Q2 = (0.012859 ,-0.000687 ,-0.001034);
P1new = (0.012712, -0.000689, -0.001095);
P2new = (0.012895, -0.000687, -0.001131);
Q1new = (0.012676, -0.000690, -0.000996);
Q2new = (0.012859, -0.000688, -0.001032);
Results when calculating self contact of 15000 line segments in a few tens of fibers, one big mess:
![]()
Anyone know of a better algorithm/corrections?
http://geomalgorithms.com/a07-_distance.html
which seems widely used and advertized on the web and works in most cases but seems to often fail horribly when line segments are nearly parallel. I've been messing with the SMALL_NUM value for division overflow to no avail. The calculated distance can still vary widely when nearly parallel.
I managed to isolate a specific incident where this happens in my code. The distance between segments P1P2 and Q1Q2 changed abruptly in one timestep from 1.05 mm to 0.90 mm (yarn radius = 1 mm), causing abrupt compression spikes. In reality the distance in the original timestep is definitely also around 0.90 mm but is not calculated as such. I find that the values of s and t (s=0 for P1, 1 for P2, t=0 for Q1, 1 for Q2) for the closest points are originally 0 and 0 (as well as in the previous time steps) and then change abruptly to 0 and around 0.29 in the new time step. What it should be, I still need to check out.
P1 = (0.012711 ,-0.000688 ,-0.001097);
P2 = (0.012895 ,-0.000686 ,-0.001133);
Q1 = (0.012676 ,-0.000689 ,-0.000999);
Q2 = (0.012859 ,-0.000687 ,-0.001034);
P1new = (0.012712, -0.000689, -0.001095);
P2new = (0.012895, -0.000687, -0.001131);
Q1new = (0.012676, -0.000690, -0.000996);
Q2new = (0.012859, -0.000688, -0.001032);
Results when calculating self contact of 15000 line segments in a few tens of fibers, one big mess:

Anyone know of a better algorithm/corrections?