1 //----------------------------------------------------------------------
2 // File: kd_fix_rad_search.cpp
3 // Programmer: Sunil Arya and David Mount
4 // Description: Standard kd-tree fixed-radius kNN search
5 // Last modified: 05/03/05 (Version 1.1)
6 //----------------------------------------------------------------------
7 // Copyright (c) 1997-2005 University of Maryland and Sunil Arya and
8 // David Mount. All Rights Reserved.
9 //
10 // This software and related documentation is part of the Approximate
11 // Nearest Neighbor Library (ANN). This software is provided under
12 // the provisions of the Lesser GNU Public License (LGPL). See the
13 // file ../ReadMe.txt for further information.
14 //
15 // The University of Maryland (U.M.) and the authors make no
16 // representations about the suitability or fitness of this software for
17 // any purpose. It is provided "as is" without express or implied
18 // warranty.
19 //----------------------------------------------------------------------
20 // History:
21 // Revision 1.1 05/03/05
22 // Initial release
23 //----------------------------------------------------------------------
24
25 #include "kd_fix_rad_search.h" // kd fixed-radius search decls
26
27 //----------------------------------------------------------------------
28 // Approximate fixed-radius k nearest neighbor search
29 // The squared radius is provided, and this procedure finds the
30 // k nearest neighbors within the radius, and returns the total
31 // number of points lying within the radius.
32 //
33 // The method used for searching the kd-tree is a variation of the
34 // nearest neighbor search used in kd_search.cpp, except that the
35 // radius of the search ball is known. We refer the reader to that
36 // file for the explanation of the recursive search procedure.
37 //----------------------------------------------------------------------
38
39 //----------------------------------------------------------------------
40 // To keep argument lists short, a number of global variables
41 // are maintained which are common to all the recursive calls.
42 // These are given below.
43 //----------------------------------------------------------------------
44
45 int ANNkdFRDim; // dimension of space
46 ANNpoint ANNkdFRQ; // query point
47 ANNdist ANNkdFRSqRad; // squared radius search bound
48 double ANNkdFRMaxErr; // max tolerable squared error
49 ANNpointArray ANNkdFRPts; // the points
50 ANNmin_k* ANNkdFRPointMK; // set of k closest points
51 int ANNkdFRPtsVisited; // total points visited
52 int ANNkdFRPtsInRange; // number of points in the range
53
54 //----------------------------------------------------------------------
55 // annkFRSearch - fixed radius search for k nearest neighbors
56 //----------------------------------------------------------------------
57
annkFRSearch(ANNpoint q,ANNdist sqRad,int k,ANNidxArray nn_idx,ANNdistArray dd,double eps)58 int ANNkd_tree::annkFRSearch(
59 ANNpoint q, // the query point
60 ANNdist sqRad, // squared radius search bound
61 int k, // number of near neighbors to return
62 ANNidxArray nn_idx, // nearest neighbor indices (returned)
63 ANNdistArray dd, // the approximate nearest neighbor
64 double eps) // the error bound
65 {
66 ANNkdFRDim = dim; // copy arguments to static equivs
67 ANNkdFRQ = q;
68 ANNkdFRSqRad = sqRad;
69 ANNkdFRPts = pts;
70 ANNkdFRPtsVisited = 0; // initialize count of points visited
71 ANNkdFRPtsInRange = 0; // ...and points in the range
72
73 ANNkdFRMaxErr = ANN_POW(1.0 + eps);
74 ANN_FLOP(2) // increment floating op count
75
76 ANNkdFRPointMK = new ANNmin_k(k); // create set for closest k points
77 // search starting at the root
78 root->ann_FR_search(annBoxDistance(q, bnd_box_lo, bnd_box_hi, dim));
79
80 for (int i = 0; i < k; i++) { // extract the k-th closest points
81 if (dd != NULL)
82 dd[i] = ANNkdFRPointMK->ith_smallest_key(i);
83 if (nn_idx != NULL)
84 nn_idx[i] = ANNkdFRPointMK->ith_smallest_info(i);
85 }
86
87 delete ANNkdFRPointMK; // deallocate closest point set
88 return ANNkdFRPtsInRange; // return final point count
89 }
90
91 //----------------------------------------------------------------------
92 // kd_split::ann_FR_search - search a splitting node
93 // Note: This routine is similar in structure to the standard kNN
94 // search. It visits the subtree that is closer to the query point
95 // first. For fixed-radius search, there is no benefit in visiting
96 // one subtree before the other, but we maintain the same basic
97 // code structure for the sake of uniformity.
98 //----------------------------------------------------------------------
99
ann_FR_search(ANNdist box_dist)100 void ANNkd_split::ann_FR_search(ANNdist box_dist)
101 {
102 // check dist calc term condition
103 if (ANNmaxPtsVisited != 0 && ANNkdFRPtsVisited > ANNmaxPtsVisited) return;
104
105 // distance to cutting plane
106 ANNcoord cut_diff = ANNkdFRQ[cut_dim] - cut_val;
107
108 if (cut_diff < 0) { // left of cutting plane
109 child[ANN_LO]->ann_FR_search(box_dist);// visit closer child first
110
111 ANNcoord box_diff = cd_bnds[ANN_LO] - ANNkdFRQ[cut_dim];
112 if (box_diff < 0) // within bounds - ignore
113 box_diff = 0;
114 // distance to further box
115 box_dist = (ANNdist) ANN_SUM(box_dist,
116 ANN_DIFF(ANN_POW(box_diff), ANN_POW(cut_diff)));
117
118 // visit further child if in range
119 if (box_dist * ANNkdFRMaxErr <= ANNkdFRSqRad)
120 child[ANN_HI]->ann_FR_search(box_dist);
121
122 }
123 else { // right of cutting plane
124 child[ANN_HI]->ann_FR_search(box_dist);// visit closer child first
125
126 ANNcoord box_diff = ANNkdFRQ[cut_dim] - cd_bnds[ANN_HI];
127 if (box_diff < 0) // within bounds - ignore
128 box_diff = 0;
129 // distance to further box
130 box_dist = (ANNdist) ANN_SUM(box_dist,
131 ANN_DIFF(ANN_POW(box_diff), ANN_POW(cut_diff)));
132
133 // visit further child if close enough
134 if (box_dist * ANNkdFRMaxErr <= ANNkdFRSqRad)
135 child[ANN_LO]->ann_FR_search(box_dist);
136
137 }
138 ANN_FLOP(13) // increment floating ops
139 ANN_SPL(1) // one more splitting node visited
140 }
141
142 //----------------------------------------------------------------------
143 // kd_leaf::ann_FR_search - search points in a leaf node
144 // Note: The unreadability of this code is the result of
145 // some fine tuning to replace indexing by pointer operations.
146 //----------------------------------------------------------------------
147
ann_FR_search(ANNdist box_dist)148 void ANNkd_leaf::ann_FR_search(ANNdist box_dist)
149 {
150 register ANNdist dist; // distance to data point
151 register ANNcoord* pp; // data coordinate pointer
152 register ANNcoord* qq; // query coordinate pointer
153 register ANNcoord t;
154 register int d;
155
156 for (int i = 0; i < n_pts; i++) { // check points in bucket
157
158 pp = ANNkdFRPts[bkt[i]]; // first coord of next data point
159 qq = ANNkdFRQ; // first coord of query point
160 dist = 0;
161
162 for(d = 0; d < ANNkdFRDim; d++) {
163 ANN_COORD(1) // one more coordinate hit
164 ANN_FLOP(5) // increment floating ops
165
166 t = *(qq++) - *(pp++); // compute length and adv coordinate
167 // exceeds dist to k-th smallest?
168 if( (dist = ANN_SUM(dist, ANN_POW(t))) > ANNkdFRSqRad) {
169 break;
170 }
171 }
172
173 if (d >= ANNkdFRDim && // among the k best?
174 (ANN_ALLOW_SELF_MATCH || dist!=0)) { // and no self-match problem
175 // add it to the list
176 ANNkdFRPointMK->insert(dist, bkt[i]);
177 ANNkdFRPtsInRange++; // increment point count
178 }
179 }
180 ANN_LEAF(1) // one more leaf node visited
181 ANN_PTS(n_pts) // increment points visited
182 ANNkdFRPtsVisited += n_pts; // increment number of points visited
183 }
184