The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B5. Beijing 2008 
684 
same way, B,’ is reduced to be P I and B 2 ’ is extended to P 2 . So 
a new coincidental section PjP 2 is built. The shortage of the 
method is that the model at the comer is too sharp to accord 
with reality. We can smooth the comer model by adding 
smoothing curve points nearby comer. 
4.1.3 3D reconstruction of crossing laneway 
Compared to the straight laneway, the 3D reconstruction of 
crossing laneway is more complicated. We present a 
constrained Delaunay modeling algorithm to solve the problem. 
Many 3D reconstruction algorithms try to reduce the data’s 
dimension to make computation easier. For example, 3D data 
for surface reconstruction will be likely projected on an 
orthogonal plane (AT, XZ, or YZ) and then triangulated as 2D 
data. According to this idea, we try to change control points on 
crossing position into AT plane and create a Delaunay triangular 
network with constrained boundary, shown as figure. 4. In this 
figure, LaneA, LaneB, LaneC and LaneD cross with each other 
in a shadow region. As to the shadow region, we firstly 
compute the surface control points according to the algorithm of 
section 4.1.2. Then, a TIN surface of junction is built by 
projecting 4 laneway’s control points on XY plane. Finally, TIN 
of LaneA, LaneB, LaneC, LandD and shadow region are put 
together into an integrated model. 
As far as coal mine are concerned, the configuration of most 
laneways is limited style. Therefore we can use automatic 
creation method. This method has advantage of fitting the quick 
modification of data and getting the right model immediately. 
The disadvantage is the laneway model is rough, the visual 
effect is not good, and the texture mapping is uncontrollable. 
Aiming to those disadvantages, the third-party modeling 
software should also be used to build laneway model. Much 
programs extrude various laneway models manual on the basis 
of centerline and section on the engineering map. This method 
has advantages of good visual effect and texture mapping. It can 
describe virtual scene in very great detail. The disadvantage is 
that the method cannot update data dynamically and the 
modeling is laborious and time-consuming. 
Figure 4. Sketch map of 4 crossing laneway 
During the actual work, we take the manual modeling and 
automatic modeling with careful consideration. As for main 
laneway, rock laneway and other fixed equipment, 3DMax 
models can be taken into account to describe a lifelike virtual 
scene. As for working face models use automatic modeling 
method. Only in this way can we get a better VR effect and a 
real dynamic underground environment. 
4.2 Collision detection and response algorithm 
Underground mining environment is labyrinthic and 
intertwining. Collision detection must be used to keep the true 
feeling during walking and interacting in the virtual laneway. 
Otherwise, the observer will penetrate through the objects and 
scene and destroy the third dimension to the whole VR program. 
Granted observer can be simplified as a sphere and the object in 
the virtual environment as triangular surface, collision detection 
is a detecting algorithm between a sphere and many triangles. 
This algorithm is universal and suitable for many cases. The 
main step to detect whether collision happens is given as 
follows: 
1) Calculate the spatial relationship between the sphere and the 
triangular plane. The relation can be divided into 3 kinds: 
sphere is in front of triangle plane, sphere is crossing the 
triangle plane and sphere is in back of triangle plane. If sphere 
and plane intersect, then continue following step. Otherwise, 
the collision does not happen. 
2) Calculate the intersection point between the sphere and the 
plane. 
3) Check whether the intersection point is inside of triangle. If 
the point is inside the triangle, the collision happens. Otherwise 
continue following step. 
4) Check whether the intersection point is on the 3 edges of 
triangle. If it is on the edge, the collision happens. Otherwise, 
collision does not happen. 
If collision between the observer and objects is detected, we 
must take measure to response collision. Here we ignore the 
complicated mechanical problems and take a simple method to 
simulate the collision. The simplest method is to stop the 
observer’s moving when collision happens. Obviously, this 
method is unsuitable because the observer will stop suddenly 
during the roaming. We take another method that we change the 
observer’s moving direction from opposite the triangular 
surface into alone the triangular surface. The algorithm is 
shown as follow. 
1) According to the normal vector, sphere radius and distance 
between sphere and plane, we can get the offset alone the object 
surface plane. 
2) Offset vector add observer’s place and moving direction. The 
observer changes the direction and moves alone the surface. 
3) Continue to detect collision. And if collision happens, repeat 
the upper 2 steps. 
4.3 Industrial system integration 
VRLane not only provides an immersive and interactive 
environment, but also closely integrates with industrial 
automation system. As far as supervisors, visitors or trainees are 
concerned, they do not need to go the dangerous underground 
mining sites. On the contrary, they can investigate underground 
safety information on a PC or a remote console. It is one of the 
most attractive features of VRLane. 
Industrial automation means integration of mechanics and 
electrics and facility automation related mining production 
process, such as coal mining, ventilation, drainage, belts, coal