1. 技术文章
  2. Collision detection between 2 “linearly” moving objects in WGS84

Collision detection between 2 “linearly” moving objects in WGS84

由 允我心安 提交于 2019-11-28 13:58:24

[Edit5] Complete reedit in case you need old sources see the revision history

As Nico Schertler pointed out checking for line to line intersection is insanity as the probability of intersecting 2 trajectories at same position and time is almost none (even when including round-off precision overlaps). Instead you should find place on each trajectory that is close enough (to collide) and both objects are there at relatively same time. Another problem is that your trajectories are not linear at all. Yes they can appear linear for shor times in booth WGS84 and Cartesian but with increasing time the trajectory bends around Earth. Also your input values units for speed are making this a bit harder so let me recapitulate normalized values I will be dealing with from now:

  1. Input

    consists of two objects. For each is known its actual position (in WGS84 [rad]) and actual speeds [m/s] but not in Cartesian space but WGS84 local axises instead. For example something like this:

    const double kmh=1.0/3.6;
const double deg=M_PI/180.0;
const double rad=180.0/M_PI;
//                      lon            lat      alt
double pos0[3]={  23.000000*deg, 48.000000*deg,2500.000000 };
double pos1[3]={  23.000000*deg, 35.000000*deg,2500.000000 };
double vel0[3]={ 100.000000*kmh,-20.000000*kmh,   0.000000*kmh };
double vel1[3]={ 100.000000*kmh, 20.000000*kmh,   0.000000*kmh };

    Beware mine coordinates are in Long,Lat,Alt order/convention !!!

  2. output

    You need to compute the time in which the two objects "collide" Additional constrains to solution can be added latter on. As mentioned before we are not searching for intersection but "closest" approach instead that suffice collision conditions (like distance is smaller then some threshold).

After some taught and testing I decided to use iterative approach in WGS84 space. That brings up some problems like how to convert speed in [m/s] in WGS84 space to [rad/s] in WGS84 space. This ratio is changing with object altitude and latitude. In reality we need to compute angle change in long and lat axises that are "precisely" equal to 1m traveled distance and then multiply the velocities by it. This can be approximated by arc-length equation:

l = dang.R

Where R is actual radius of angular movement, ang is the angle change and l is traveled distance so when l=1.0 then:

dang = 1.0/R

If we have Cartesian position x,y,z (z is Earth rotation axis) then:

Rlon = sqrt (x*x + y*y)
Rlat = sqrt (x*x + y*y + z*z)

Now we can iterate positions with time which can be used to approximate closest approach time. We need to limit the max time step however so we do not miss to much of the Earth curvature. This limit is dependent on used speeds and target precision. So here the algorithm to find the approach:

  1. init

    set initial time step to the upper limit like dt=1000.0 and compute actual positions of booth objects in Cartesian space. From that compute their distance d1.

  2. iteration

    set d0=d1 then compute actual speeds in WGS84 for actual positions and add speed*dt to each objects actual WGS84 position. Now just compute actual positions in Cartesian space and compute their distance d1

    if d0>d1 then it menas we are closing to the closest approach so goto #2 again.
    In case d0==d1 the trajectories are parallel so return approach time t=0.0
    In case d0<d1 we already crossed the closest approach so set dt = -0.1*dt and if dt>=desired_accuracy goto #2 otherwise stop.

  3. recover best t

    After the iteration in #2 we should recover the best time back so return t+10.0*dt;

Now we have closest approach time t. Beware it can be negative (if the objects are going away from each other). Now you can add your constrains like 

if (d0<_max_d)
 if ((t>=0.0)&&(t<=_max_T))
  return collision ...

Here C++ source for this:

//---------------------------------------------------------------------------
#include <math.h>
//---------------------------------------------------------------------------
const double kmh=1.0/3.6;
const double deg=M_PI/180.0;
const double rad=180.0/M_PI;
const double  _earth_a=6378137.00000;   // [m] WGS84 equator radius
const double  _earth_b=6356752.31414;   // [m] WGS84 epolar radius
const double  _earth_e=8.1819190842622e-2; //  WGS84 eccentricity
const double  _earth_ee=_earth_e*_earth_e;
//--------------------------------------------------------------------------
const double _max_d=2500.0;         // [m] collision gap
const double _max_T=3600000.0;      // [s] max collision time
const double _max_dt=1000.0;        // [s] max iteration time step (for preserving accuracy)
//--------------------------------------------------------------------------
//                      lon            lat      alt
double pos0[3]={  23.000000*deg, 48.000000*deg,2500.000000 }; // [rad,rad,m]
double pos1[3]={  23.000000*deg, 35.000000*deg,2500.000000 }; // [rad,rad,m]
double vel0[3]={ 100.000000*kmh,-20.000000*kmh,   0.000000*kmh }; // [m/s,m/s,m/s]
double vel1[3]={ 100.000000*kmh,+20.000000*kmh,   0.000000*kmh }; // [m/s,m/s,m/s]
//---------------------------------------------------------------------------
double divide(double x,double y)
        {
        if ((y>=-1e-30)&&(y<=+1e-30)) return 0.0;
        return x/y;
        }
void  vector_copy(double *c,double *a)         { for(int i=0;i<3;i++) c[i]=a[i];       }
double vector_len(double *a) { return sqrt((a[0]*a[0])+(a[1]*a[1])+(a[2]*a[2])); }
void  vector_len(double *c,double *a,double l)
        {
        l=divide(l,sqrt((a[0]*a[0])+(a[1]*a[1])+(a[2]*a[2])));
        c[0]=a[0]*l;
        c[1]=a[1]*l;
        c[2]=a[2]*l;
        }
void  vector_sub(double *c,double *a,double *b) { for(int i=0;i<3;i++) c[i]=a[i]-b[i]; }
//---------------------------------------------------------------------------
void WGS84toXYZ(double *xyz,double *abh)
    {
    double  a,b,h,l,c,s;
    a=abh[0];
    b=abh[1];
    h=abh[2];
    c=cos(b);
    s=sin(b);
    // WGS84 from eccentricity
    l=_earth_a/sqrt(1.0-(_earth_ee*s*s));
    xyz[0]=(l+h)*c*cos(a);
    xyz[1]=(l+h)*c*sin(a);
    xyz[2]=(((1.0-_earth_ee)*l)+h)*s;
    }
//---------------------------------------------------------------------------
void WGS84_m2rad(double &da,double &db,double *abh)
    {
    // WGS84 from eccentricity
    double p[3],rr;
    WGS84toXYZ(p,abh);
    rr=(p[0]*p[0])+(p[1]*p[1]);
    da=divide(1.0,sqrt(rr));
    rr+=p[2]*p[2];
    db=divide(1.0,sqrt(rr));
    }
//---------------------------------------------------------------------------
double collision(double *pos0,double *vel0,double *pos1,double *vel1)
    {
    int e,i,n;
    double p0[3],p1[3],q0[3],q1[3],da,db,dt,t,d0,d1,x,y,z;
    vector_copy(p0,pos0);
    vector_copy(p1,pos1);
    // find closest d1[m] approach in time t[sec]
    dt=_max_dt; // [sec] initial time step (accuracy = dt/10^(n-1)
    n=6;        // acuracy loops
    for (t=0.0,i=0;i<n;i++)
     for (e=0;;e=1)
        {
        d0=d1;
        // compute xyz distance
        WGS84toXYZ(q0,p0);
        WGS84toXYZ(q1,p1);
        vector_sub(q0,q0,q1);
        d1=vector_len(q0);
        // nearest approach crossed?
        if (e)
            {
            if (d0<d1){ dt*=-0.1; break; }                  // crossing trajectories
            if (fabs(d0-d1)<=1e-10) { i=n; t=0.0; break; }  // parallel trajectories
            }
        // apply time step
        t+=dt;
        WGS84_m2rad(da,db,p0);
        p0[0]+=vel0[0]*dt*da;
        p0[1]+=vel0[1]*dt*db;
        p0[2]+=vel0[2]*dt;
        WGS84_m2rad(da,db,p1);
        p1[0]+=vel1[0]*dt*da;
        p1[1]+=vel1[1]*dt*db;
        p1[2]+=vel1[2]*dt;
        }
    t+=10.0*dt; // recover original t
//  if ((d0<_max_d)&&(t>=0.0)&&(t<=_max_T)) return collision; else return no_collision;
    return t;
    }
//---------------------------------------------------------------------------

Here an overview of example:

Red is object0 and Green is object1. The White squares represents position at computed collision at time t_coll [s] with distance d_coll [m]. Yellow squares are positions at user defined time t_anim [s] with distance d_anim [m] which is controlled by User for debugging purposes. As you can see this approach works also for times like 36 hours ...

Hope I did not forget to copy something (if yes comment me and I will add it)

You do not show any code, so I will just give the main ideas and leave the coding to you. Come back if you try some code and are stuck, but show your effort and your code so far.

There are multiple ways to solve your problem. One way is to set parametric equations for each object, giving your two functions in time t. Set the results of those functions equal and solve for time. For 3D coordinates that gives you three questions, one for each coordinate, and it is very unlikely that the values of t will be the same for all three equations. If they are the same, that is the time of your collision.

Another way, which allows for some floating-point rounding errors, is to change the frame of reference to that of one of the objects. You subtract the two velocity vectors, say v2-v1, and you now have the velocity of the second object relative to the first object. Now find the distance from the now-stationary first object to the line of the moving second object. If you don't know how to do that, look up 'distance from point to line' in your favorite search engine. You then see if that distance is small enough for you to consider it as a collision--you are unlikely to get a perfect collision, a zero distance, given floating-point rounding errors. If it is small enough, you then see if that collision is reached in the future or was reached in the past. You may want to find the projection of the point on the line as an intermediate value for that last calculation.

Is that clear?

标签
c++
math
vector
3d
wgs84
易学教程内所有资源均来自网络或用户发布的内容，如有违反法律规定的内容欢迎反馈！
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!