Exemple #1
0
        private void ransac_btn_Click(object sender, EventArgs e)
        {
            ransac = ApplyRansac();
            //we need to sort the points, otherwise we will get a random figure
            List <Point> sorted1 = new List <Point>();
            List <Point> sorted2 = new List <Point>();

            for (int j = 0; j < Shape1.Count; j++)
            {
                if (ransac.best_consensus1.Contains(Shape1[j]))
                {
                    sorted1.Add(Shape1[j]);
                    sorted2.Add(Shape2[j]);
                }
            }

            List <Point> new_Shape2 = ApplyTransformation(ransac.best_model, sorted2);
            Pen          pBlue      = new Pen(Brushes.Blue, 1);
            Pen          pRed       = new Pen(Brushes.Red, 1);
            Graphics     g          = ransac_panel.CreateGraphics();

            DisplayShape(sorted1, pBlue, g);
            DisplayShape(new_Shape2, pRed, g);
            MessageBox.Show("Cost = " + ransac.best_error);
        }
Exemple #2
0
        Ransac_Algo ApplyRansac()
        {
            Ransac_Algo rans = new Ransac_Algo();
            int         k    = 20; //iterations for the algorithm
            int         t    = 7;  //treshold to determine if we accept the new point or not, if it adds less than 7 to the cost it is valid
            int         n    = 3;  //number of initial random point
            int         d    = 4;  //minimum data points to accept the model as valid

            int iterations = 0;    //counter of iterations done

            /*
             * Transformation best_model = new Transformation();
             * List<Point> best_consensus_set1 = new List<Point>();
             * List<Point> best_consensus_set2 = new List<Point>();
             * Ransac_Algo rans = new Ransac_Algo();
             */

            double best_error = int.MaxValue;
            Random rand       = new Random(); //used to select random indexes


            while (iterations < k)
            {
                List <int>     random_index    = new List <int>(); //list that will contain random indexes from shapes
                List <Point>   maybe_inliners1 = new List <Point>();
                List <Point>   maybe_inliners2 = new List <Point>();
                Transformation maybe_model     = new Transformation();
                double         maybe_cost      = 0; //cost from the random points

                //selecting random points (using indexes) from data
                for (int i = 0; i < n; i++)
                {
                    int index = rand.Next(Shape1.Count);
                    while (random_index.Contains(index))
                    {
                        index = rand.Next(Shape1.Count);
                    }
                    random_index.Add(index);
                    maybe_inliners1.Add(Shape1[index]);
                    maybe_inliners2.Add(Shape2[index]);
                }

                maybe_model = ICPTransformation.ComputeTransformation(maybe_inliners1, maybe_inliners2);    //computing model for random chosen points
                maybe_cost  = ICPTransformation.ComputeCost(maybe_inliners1, maybe_inliners2, maybe_model); //computing cost for random chosen points
                List <Point> consensus_set1 = new List <Point>(maybe_inliners1);
                List <Point> consensus_set2 = new List <Point>(maybe_inliners2);

                for (int i = 0; i < Shape1.Count; i++)
                {
                    if (!random_index.Contains(i))
                    {
                        maybe_inliners1.Add(Shape1[i]);
                        maybe_inliners2.Add(Shape2[i]);
                        consensus_set1.Add(Shape1[i]);
                        consensus_set2.Add(Shape2[i]);
                        Transformation temp_model = ICPTransformation.ComputeTransformation(consensus_set1, consensus_set2);
                        double         temp_cost  = ICPTransformation.ComputeCost(consensus_set1, consensus_set2, temp_model);
                        double         dif        = Math.Abs(maybe_cost - temp_cost);
                        if (dif >= t)
                        {
                            consensus_set1.RemoveAt(consensus_set1.Count - 1);
                            consensus_set2.RemoveAt(consensus_set2.Count - 1);
                        }
                    }
                }

                if (consensus_set1.Count > d) //checking if consensus has enough points to make the model valid
                {
                    Transformation better_model = ICPTransformation.ComputeTransformation(consensus_set1, consensus_set2);
                    double         this_error   = ICPTransformation.ComputeCost(consensus_set1, consensus_set2, better_model);


                    if (this_error < best_error)
                    {
                        best_error           = this_error;
                        rans.best_model      = better_model;
                        rans.best_consensus1 = consensus_set1;
                        rans.best_consensus2 = consensus_set2;
                        rans.best_error      = this_error;
                    }
                }

                iterations++;
            }
            return(rans);
        }