JP5781045B2 - Object search apparatus and method - Google Patents

Description

  The present embodiment relates to an object search apparatus and method, and relates to a method for managing a three-dimensional (3D) object of an application such as GIS or CAD.

  The 3D database manages a set of objects arranged in space-time. Receives a query that specifies a range and returns a list of objects in that range. R-tree, quad-tree and their extensions are used as the base data structure / index.

R-tree (Antonin Guttman: R-Trees: A Dynamic Index Structure for Spatial Searching, Proc. 1984 ACM SIGMOD International Conference on Management of Data, pp. 47-57. ISBN 0-89791-128-8) Quad-tree (Raphael Finkel and J.L. Bentley (1974). “Quad Trees: A Data Structure for Retrieval on Composite Keys”. Acta Informatica 4 (1): 1? 9. Doi: 10.1007 / BF00288933)

  When there is another arbitrary object between a certain viewpoint and a certain object, the object is said to be blocked. A method of obtaining a list of obstructed objects by specifying an object group and a viewpoint is known. A query for an object that exists in a semi-linear range from the object toward the viewpoint is executed for each object, and if there is even one object other than itself within that range, the object is blocked Judge.

  Since each of these processes is performed based on R-tree or Quad-tree, high performance is achieved with a single search. However, since the search is performed for all objects, there is a problem that the search becomes slow.

  Such a large number of uniform searches can be done in a 3D database that manages the arrangement of PV (photovoltaic power generation) devices, a list of objects (PV devices) that fall into the shadow of the building, and a view from a certain viewpoint. Queries that require a list of objects to enter frequently occur.

  On the other hand, 3D drawing technology is being researched, developed and utilized in CAD and games. This processing can be executed at a very high speed by using dedicated hardware such as a GPU, and this hardware is widely used and is inexpensive. Therefore, it is conceivable to generate drawing data by 3D drawing from the viewpoint direction and search for an obstructed object based on the drawing data. However, this search is fast, but the accuracy is not high.

  An object of one aspect of the present invention is to quickly search for an object obstructed by another object from a certain viewpoint in space.

  An object search apparatus according to an aspect of the present invention includes a database, a rendering unit, a calculation unit, and an object search unit.

  The database stores arrangement information of a plurality of objects in the space.

  The rendering unit draws the plurality of objects from a viewpoint set in the space.

  The calculation unit performs drawing using the rendering unit, and calculates the number of objects in which at least some images are hidden by images of other objects.

  The object search unit uses the number calculated by the calculation unit as an upper limit value of the number of search results, and searches for an object in the space that is at least partially obstructed by another object when viewed from the viewpoint Based on the database.

The block diagram of the object search device concerning this embodiment. The flowchart of an online search result estimation process. The flowchart of the whole object search apparatus. The figure of the object arrangement | positioning in Example A. The figure which shows the tree structure showing the object arrangement in execution example A. The flowchart of a comparison frequency reduction search process. The figure which shows the table of the comparison frequency / cache miss frequency in execution example A. FIG. 10 is a diagram illustrating object arrangement in the embodiment B. The figure which shows the tree structure showing the object arrangement in execution example B. The flowchart of a cache miss reduction search process. The figure which shows the table of the comparison frequency / cache miss frequency in execution example B.

  Hereinafter, this embodiment will be described in detail with reference to the drawings.

  FIG. 1 shows an object search apparatus for a 3D database according to an embodiment of the present invention.

  The object hierarchical structure forming unit 101 calculates a minimum bounded box (MBB) for each of a plurality of objects arranged in the space. Each side of the minimum enclosing cuboid to be calculated is parallel to any of the x-axis, y-axis, and z-axis. Then, the objects close to each other are grouped (grouped), and the MBB of the group is calculated. A hierarchical structure expressed by a balanced tree structure is expressed by using MBB of each object as a leaf, a cluster as a node, and a cluster as a node. In addition, when an object is inserted or deleted, the tree structure is changed so as to maintain the characteristics. The tree structure includes information on a parent-child relationship of each node and an arrangement position of each object in the MBB space.

  The tree structure generation method described here is R-tree (Antonin Guttman: R-Trees: A Dynamic Index Structure for Spatial Searching, Proc. 1984 ACM SIGMOD International Conference on Management of Data, pp. 47-57. ISBN 0- 89791-128-8) and Quad-tree, so refer to the relevant document for details.

  The object MBB tree storage unit 102 is a database that stores the hierarchical structure generated by the object hierarchical structure forming unit 101. In this embodiment, the MBB of the input object and the MBB of the grouped object are calculated, but if the input object is parallel to each side x-axis, y-axis, and z-axis of the rectangular parallelepiped, the object No MBB calculation is required. In this embodiment, an object is handled by its MBB, but this embodiment is not limited to this.

  In the processing of the rendering unit 104, the search result number estimation unit (calculation unit) 103, and the object search unit 105 described below, the object processing is performed by the MBB of the object unless otherwise specified.

  The rendering unit 104 reads out a list of objects arranged in the 3D space from the object MBB tree storage unit 102. The rendering unit 104 performs 3D rendering processing of each object in space from the viewpoint direction given by the query. Information obtained during the rendering process or the result of the rendering process (drawing data) is used by the search result number estimation unit 103. There is no need to actually display the information obtained during the process or the result of the rendering process on the display device.

  The search result number estimation unit 103 receives a query representing a viewpoint, and uses the rendering unit 104 to 3D render each object from the viewpoint indicated by the query. By using information obtained in the middle of rendering or processing results of rendering, an upper limit value of the number of objects obstructed by other objects (upper limit value of the number of search results) is estimated.

  Here, as the query representing the viewpoint, 3D coordinates representing the position (perspective camera position of perspective projection), 3D vector representing the direction of the line of sight (considering that the viewpoint is at infinity, the line of sight of the parallel projection), or equivalent A data structure having the following information is used. Here, the subject of the viewpoint is assumed to be an omnidirectional camera. An example of the use of this embodiment is to obtain a list of objects (objects obstructed by other objects) that cause a shadow by placing a point light source at the position.

  Estimating the upper limit value of the number of search results using the rendering process result is called “offline search result estimation process”. On the other hand, estimating the upper limit of the number of search results using information obtained during the rendering process is called “online search result estimation process”. Both processes give the same result.

  In the “offline search result estimation process”, the drawing data of the rendering result is used to extract only the object drawn in a complete form by determination. That is, all objects that are not hidden by the images of other objects (not overlapping the images of other objects) are specified. The specified set of objects is a set of objects that are not hidden by images of other objects. When a set of objects hidden by images of other objects is desired, it can be obtained by taking the complement of the set from all sets. In the present embodiment, the latter set of objects hidden by images of other objects is acquired. An object hidden by an image of another object can be a candidate for an object that is actually blocked by the other object in the 3D space.

  Here, as a specific determination method, individual objects are rendered independently independently, and the images of the individual rendering results and the images of the individual objects included in the rendering results of all the above-mentioned objects are the same. Determine. If they are the same, the object can be determined as an object that is not hidden, and if not, it can be determined as an object that is hidden.

  Next, “online search result estimation processing” will be described.

  FIG. 2 shows a flowchart of the online search result estimation process.

  First, the list is initialized (S201). Sequentially draws a set of objects arranged in 3D space. At this time, each object is drawn one by one in order (S202). In the drawing of the object, the following processing is performed for each pixel (S203).

  That is, if another object has already been drawn on the pixel to be drawn (YES in S204), it is determined whether the object to be drawn (current object) is in front of the already drawn object. (S205). If it is in front (NO in S205), it is determined that the existing object is hidden by the currently drawn object, and the identifier of the existing object is added to the list (S206). The pixel corresponding to the existing object is overwritten with the data (S207). If not (YES in S205), that is, if the current object is behind the already drawn object, it is determined that the current object is hidden by the already drawn object, and the current object The object identifier is added to the list (S208), and drawing is skipped. If nothing is drawn on the pixel to be drawn (NO in S204), the object is drawn (S207).

  Finally, a list is output (S209). The number of objects included in the list is an estimation result of the upper limit value of the number of objects blocked by other objects.

  Here, both “offline search result estimation” and “online search result estimation” are completely dependent on the resolution of drawing (meaning how much larger or smaller than actual drawing). Accurate accuracy cannot be obtained. Depending on the resolution, it is determined that it is hidden by drawing although it is not actually blocked, or it is determined that it is actually hidden but not hidden by drawing, but the drawing resolution is adjusted. This can be adjusted to either.

  In this embodiment, by setting the resolution lower (by drawing the object larger than the actual size), even if it is not obstructed by other objects in the actual space, Allow to be determined to be hidden. However, although it is obstructed by another object in the actual space, it is not determined that it is not hidden by the image of the other object in the drawing. As described above, the above-mentioned “offline search result estimation” or “online search result estimation” can obtain the upper limit of the number of search results when searching for an object that is actually blocked. . By performing a search using the upper limit value, an efficient search is possible as will be described later.

  The object search unit 105 selects an appropriate search strategy from a plurality of search strategies given in advance based on the estimation result (upper limit value) of the search result number estimation unit 103. Then, according to the selected search strategy, the tree is read from the object MBB tree storage unit 102 and searched.

  Many specific search strategies are conceivable and are not limited to specific ones. In this embodiment, as an example, a comparison frequency reduction strategy (second search strategy) and a cache miss reduction strategy (first search strategy) are introduced. The comparison frequency reduction strategy is a search strategy that aims to reduce the number of comparisons (number of determinations) by performing recursive determination as necessary. The cache miss reduction strategy is a search strategy aiming to reduce the number of times of reading from the object MBB tree storage unit 102 (number of cache misses) by performing all recursive determinations when a node arrives during search.

  FIG. 3 is a flowchart showing the overall operation flow.

  The search result number estimation unit 103 uses the rendering unit 104 to estimate the upper limit of the number of objects obstructed by other objects (S101). The rendering unit 104 may perform either search result estimation processing, offline or online.

  The object search unit 105 compares the estimated number (upper limit value) with a threshold value (S102), and the search strategy (search strategy 1, search strategy 2) differs depending on whether the upper limit value is less than the threshold value or greater than the threshold value. And a space (tree) is searched with the selected search strategy (S103, S104). By searching, an object that is actually blocked by another object is detected. The search result output unit 106 outputs a list of the objects (S105).

  Hereinafter, the operation of the present embodiment will be described in detail as Examples A and B based on specific examples.

(Example A)
In the present embodiment, it is assumed that the object set as shown in FIG. 4 is searched for an object that is blocked by another object when viewed from a certain viewpoint.

  The search result number estimation unit 103 performs offline search result estimation.

  The object search unit 105 selects and executes either the comparison frequency reduction strategy or the cache miss reduction strategy. Specifically, when the estimation result of the search result estimation unit 103 is 3 or less, the comparison number reduction strategy is selected, and when it is larger than 3, the cache miss reduction strategy is selected. The number 3 here is just an example, and can be set to any value.

(Operation of Object Hierarchy Configuration Unit 101 in Example A)
A tree structure is constructed by inputting an object group and hierarchically grouping close objects. In this example, it is assumed that the tree structure shown in FIG. 5 is obtained. The leaf node of the tree structure is an MBB of the object, and the other nodes are nodes representing the MBB of the grouped objects.

  A leaf node with fine pitch hatching is a leaf node that represents an object that matches the query (ie, an object that is blocked by another object).

  Further, a hatched frame with a rough pitch represents a unit of reading from the object MBB tree storage unit 102. For example, node a and node b are read simultaneously. Node 4 and node 5 are read simultaneously. For the sake of explanation, it is assumed that only one read unit data can be stored in the cache memory, and when another read unit data is read, the existing read unit is overwritten. For example, when nodes 10 and 11 are read in a state where nodes 4 and 5 exist in the cache, nodes 4 and 5 are overwritten. Actually, if the size of the cache is large, there is a high possibility that the previous data remains in the cache even if other data is read. However, in this embodiment, for the sake of simplicity, a simple model in which only data for one reading unit at a time is stored in the cache will be described.

(Operation of Object MBB Tree Storage Unit 102 in Example A)
The object MBB tree storage unit 102 stores a tree structure as shown in FIG.

(Operation of the rendering unit 104 in the embodiment A)
The rendering unit 104 draws (renders) and outputs an image as shown in FIG.

(Operation of the offline search result number estimation unit 103 in Example A)
The offline search result number estimation unit 103 receives a query representing a viewpoint and performs rendering using the rendering unit 104. Thereby, a rendering result (drawing data) as shown in FIG. 4 is obtained. In FIG. 4, only the image of the object f is partially hidden, and all the images of the objects other than the object f are completely visible. In other words, it can be estimated that there is at most one obstructed object.

(Operation of Object Search Unit 105 in Example A)
The estimation result (upper limit value) of the offline search result number estimation unit 103 is compared with the threshold value. Here, the threshold is 3. If the estimation result is 3 or less, the comparison number reduction strategy is used, and if it is larger than 3, the cache miss reduction strategy is used. Since the estimation result is 1, which is 3 or less, the comparison number reduction search strategy is adopted here.

  FIG. 6 shows a flowchart of the comparison number reduction search process.

  For the sake of simplicity, it is assumed that all nodes do not exist in the cache at the start of the search.

  The object search unit 105 first loads the node 1 into the cache (S301), and includes an object in which the node 1 may satisfy the query, that is, the MBB of the node 1 is blocked by other objects. It is compared and determined whether or not there is a possibility (S303). Note that step S302 is omitted for node 1, which is the root node. Here, it is determined that there is a possibility that the MBB of the node 1 includes an object blocked by another object (YES in step S303). When node 1 is loaded, a cache miss occurs and reading from the object MBB tree storage unit 102 occurs. Here, an example of the comparison / determination method is shown.

  In general, there is already a technique that uses a spatial search tree to determine whether there is another object between the line segment connecting the viewpoint (such as the sun) from the object (if there is another object, the object will become a shadow. Determined to be). In other words, it is possible to determine whether or not other objects exist between the line segments connecting each vertex of MBB of the node to the sun. If any other object is found, it is determined that the node may become a shadow, so it may be possible to satisfy the query. Conversely, if no object is found, it is determined that there is no possibility of satisfying the query. In this embodiment, first, if at least one of the line segments obtained by connecting the eight vertices of the MBB of the node 1 and the viewpoint (such as the sun) collides with any of the objects a to h, the node 1 determines that there is a possibility of satisfying the query (possibly including an object blocked by other objects). The comparison / determination may be performed using methods other than those described here.

  Next, the flow proceeds to a recursive search (S304) in which this flow is recursively, and the nodes 2 and 3 that are child nodes of the node 1 are loaded into the cache (S301). That is, a cache miss occurs, and reading of nodes 2 and 3 from the object MBB tree storage unit 102 occurs. Compare / determine whether node 2 and node 3 may satisfy the query. Here, for the comparison frequency reduction search, both of the two nodes are not compared / determined at the same time, but only one of them arbitrarily selected is compared / determined (S302, S303). Here, it is assumed that node 3 is selected. Here, it is determined that the node 3 may satisfy the query (YES in S303). For node 2, at this time point, only the information that comparison / determination is not performed and comparison / determination is not performed is stored.

  Next, the process proceeds to a recursive search (S304), and the nodes 6 and 7 are loaded into the cache (S301). That is, a cache miss occurs, and the nodes 6 and 7 are read from the object MBB tree storage unit 102. Here, node 6 is selected. Here, it is determined that the node 6 may satisfy the query (YES in S302 and S303). As for the node 7, at this time point, only information that comparison / determination is not performed and comparison / determination is not performed is stored.

  Next, the process proceeds to recursive search (S304), and the nodes 12 and 13 are loaded into the cache (S301). That is, a cache miss occurs, and the nodes 12 and 13 are read from the object MBB tree storage unit 102. Here, it is assumed that the node 12 is selected. Node 12 is a leaf node. It is determined whether the object f of the node 12 is blocked by another object. That is, it is determined whether another object exists on a line segment connecting each vertex of the MBB of the object f and the viewpoint (sun etc.). Here, it is determined that the object e exists. That is, it is determined that the object f is blocked by the object e. Here, the value (upper limit value) of the estimation result is 1. Since one occluded object has been found, it can be seen that no more occluded objects can be found. Thus, the search ends. If node 13 is selected before node 12, it is determined that node 13 does not satisfy the query, then sibling node 12 is selected, and it is determined that the query is satisfied as described above. And the search ends.

(Effect of Example A)
As shown in FIG. 7, in the comparison number reduction search of this example, the number of comparisons was 4, and the number of cache misses was 4.

  If we consider a cache miss reduction search instead of a comparison count reduction search, node 1, node 2, node 3, node 6, node 7, node 12, and node 13, a total of 7 times A comparison is made and the number of cache misses is the same four times (see FIG. 7). If the number of cache misses is the same, it is more efficient to have a smaller number of comparisons. Thus, the present embodiment can finish the search with a small number of comparisons. The operation of the cache miss reduction search will be described later.

  However, this is an evaluation when all optimal selections are possible in nondeterministic selection of nodes (such as selection of node 2 and node 3), so the number of comparisons is actually more than 4 times. there is a possibility. However, the upper limit is 7 times as in the cache miss reduction search, and the cache miss reduction search always performs 7 comparisons, so that an efficient selection can be made almost always.

(Example B)
In the present embodiment, it is assumed that an object obstructed by another object is searched for the set of objects as shown in FIG.

  The search result number estimation unit 103 performs online search result estimation.

  The object search unit 105 selects and executes either the comparison frequency reduction strategy or the cache miss reduction strategy. Specifically, when the estimation result of the search result estimation unit 103 is 3 or less, the comparison frequency reduction strategy is selected, and when it is larger than 3, the cache miss reduction strategy is selected.

(Operation of Object Hierarchy Configuration Unit 101 in Example B)
A tree structure is constructed by inputting an object group and hierarchically grouping close objects. The leaf node of the tree structure is the MBB of the object, and the other nodes represent the MBB of the grouped objects. In this example, it is assumed that the tree structure shown in FIG. 9 is obtained. A leaf node to which hatching with fine pitch is applied is a leaf node representing an object matching the query.

  Further, a hatched frame with a rough pitch represents a unit of reading from the object MBB tree storage unit 102. As in the case of the embodiment A, the cache memory will be described with a simple model in which only one piece of read unit data can be stored.

(Operation of Object MBB Tree Storage Unit 102 in Example B)
The object MBB tree storage unit 102 stores a tree structure as shown in FIG.

(Operations of the rendering unit 104 and the online search result number estimation unit 103 in the embodiment B)
The online search result number estimation unit 103 receives a query representing a viewpoint, and the rendering unit 104 performs rendering (rendering).

  The rendering unit 104 sequentially draws all the objects according to the operation flow shown in FIG. Here, it is assumed that objects a to h are drawn in this order. The final drawing result is as shown in FIG.

  First, a list for storing identifiers of objects hidden by images of other objects is initialized as empty (S201).

  Draw object a against the sky background. Since no other object exists in the drawing area, the object a is drawn as it is (NO in S204, S207).

  Similarly, the object b is drawn (NO in S204, S207).

  Next, the object c is drawn, and the object b is already drawn on some pixels for which the object c is to be drawn. The object c that is currently being drawn is in the foreground. Therefore, the identifier of the drawn object b is added to the list (NO in S204, S206, S207).

  Next, the objects d and e are drawn. Since both are to be drawn in an area where no other object is drawn, the drawing is performed as it is (NO in S204, S207).

  Next, the object f is drawn, and the object e has been drawn on some pixels for which the object f is to be drawn. And e already drawn is in the foreground. Therefore, the object e to be drawn is added to the list (YES in S204, YES in S205, S208).

  Similarly, it can be seen that the objects g and h have other objects in front. Therefore, the objects g and h that are currently drawn are added to the list (YES in S204, YES in S205, and S208).

  When the above is completed, a list is output (S209). The length of the list is 4. Therefore, the estimated number (upper limit) is 4.

(Operation of Object Search Unit 105 in Example B)
Since the result of estimating the number of offline search results is not 3 or less, here, the search for reducing the number of cache misses is adopted.

  FIG. 10 shows a flowchart of the cache miss reduction search process.

  At the start of the search, for the sake of simplicity, it is assumed that all nodes do not exist in the cache.

  In the comparison frequency reduction search process, the node 1 is first cached (S401). That is, a cache miss occurs and the data of node 1 is read from the object MBB tree storage unit 102 to the CPU. An array for storing the determination result is initialized (S402). Step S403 skips for node 1, which is the root node. It is compared and determined whether or not there is a possibility that the node 1 satisfies the query (S404). Here, it is determined that there is a possibility (should be recursive), and the determination result is stored in the array (S404). Step S405 skips for node 1, which is the root node. In step S406, since the determination result that node 1 should be recursed is recorded in the array, the process proceeds to a recursive search (S407) that recurs this flow.

  Next, node 2 and node 3 which are child nodes of node 1 are loaded into the cache, and the arrays for nodes 2 and 3 are initialized (S401, S402). That is, a cache miss occurs and reading from the object MBB tree storage unit 102 occurs. It is compared and determined whether or not the nodes 2 and 3 may satisfy the query (S403, S404). In the comparison frequency reduction search described above, only comparison / determination is performed for only one of the nodes. In this cache miss frequency reduction search, both comparisons / determinations are performed. It is determined that there is a possibility that both the nodes 2 and 3 satisfy the query, and the determination result is stored in the array (S404). In step S406, since the determination result that both nodes 2 and 3 should be recursed is recorded, either one of the nodes is selected and a recursive search is performed. Here, node 2 is selected, and a recursive search is performed for node 2 (that is, the tree descends from node 2) (S405, S406, S407).

  Next, processing similar to that of nodes 2 and 3 is performed for nodes 4 and 5 (S401, S402, S403, S404, S405, and S406). Also in this case, a cache miss occurs, and reading from the object MBB tree storage unit 102 occurs. The determination result that nodes 4 and 5 should be recursed is recorded in the array for nodes 4 and 5. Here, node 4 is selected and a recursive search is performed.

  Next, node 8 and node 9 which are child nodes of node 4 are loaded into the cache, and an array for nodes 8 and 9 is initialized (S401, S402). That is, a cache miss occurs and reading from the object MBB tree storage unit 102 occurs. It is compared and determined whether or not the nodes 8 and 9 may satisfy the query (S403, S404). Node 8 and node 9 are leaf nodes. Node 9 satisfies the query, which finds object b. Node 9 is determined not to satisfy the query. Since both nodes 8 and 9 are leaf nodes, a determination result indicating that they should not be recurred is recorded in the array. This completes the search up to the leaf node, but the estimated number (upper limit) has not yet been reached. For this reason, the search must continue. Therefore, the process returns to node 4 and node 5 (NO in step S406).

  Nodes 4 and 5 select a different child node (that is, descend the tree from node 5). At this time, if the comparison number reduction search is performed, in order to determine whether or not the node 5 should be retired (recursion should be performed) at this time, a cache miss occurs and the object MBB tree storage unit 102 again receives the node 4, 5 must be read. On the other hand, in the cache miss frequency reduction search according to the present embodiment, whether or not the node 5 should be recursed has already been determined (determination that the recursion should be performed). Therefore, a cache miss (reading from the object MBB tree storage unit 102) does not occur. As a result, the number of cache misses is reduced as compared with the comparison number reduction search. Here, the tree descends from node 5 (S407).

  Next, processing similar to that of the nodes 8 and 9 is performed for the nodes 10 and 11 that are the child nodes of the node 5 (NO in S406 of S401 to S405). At this time, a cache miss occurs and the nodes 10 and 11 are read. Since both the nodes 10 and 11 are leaf nodes, the determination result that should not be recursive is recorded in the array. Node 10 and node 11 are leaf nodes, but neither node satisfies the query.

  Next, the process returns to the node 4 and the node 5, but since there are no further child nodes to descend, the process further returns to the node 2 and the node 3. For node 3, the determination result that recursion should be recorded in the array for nodes 2 and 3, this time node 3 is selected (down the tree from node 3).

  Next, processing similar to that of nodes 4 and 5 is also performed for nodes 6 and 7 that are child nodes of node 3 (YES in S406 of S401 to S405). At this time, a cache miss occurs, and reading from the object MBB tree storage unit 102 occurs. In the arrays for nodes 6 and 7, a determination result indicating that both should be recursed is recorded. Here, it is assumed that the tree descends from the node 6 (S407).

  Next, processing similar to that of the nodes 8 and 9 is performed for the nodes 12 and 13 (NO in S401 to S405 and S406). A cache miss occurs and the nodes 12 and 13 are read. Since both the nodes 12 and 13 are leaf nodes, the determination result that should not be recursed is recorded in the array. Nodes 12 and 13 are leaf nodes. Since node 13 satisfies the query, object f is found. Note that the node 12 does not satisfy the query. Returning to the nodes 6 and 7, the tree descends from the node 7 (S407).

  Next, processing similar to that of the nodes 12 and 13 is performed for the nodes 14 and 15 which are child nodes of the node 7 (NO in S401 to S405 and S406). A cache miss occurs and the nodes 14 and 15 are read. Nodes 14 and 15 are leaf nodes. Since both satisfy the query, objects g and h are found. Since the same number as the estimated number (upper limit value) has been found, the search is terminated.

(Effect of Example B)
In this example, as shown in FIG. 11, the number of comparisons was 15 and the number of cache misses was 8.

  If a case where a search for reducing the number of comparisons is adopted instead of a cache miss reduction search is considered, a cache miss occurs every time the upper node is returned (as described above, for simplicity, in this embodiment, (Since it is explained in the simple model that only one read unit (sibling node group) is in the cache at the same time), a total of 15 comparisons are made and the number of cache misses is 11 (see Figure 11) ). If the number of comparisons is the same, it is more efficient that the number of cache misses is smaller.

  In this way, in this embodiment, the search can be completed with a small number of cache misses (number of readings).

(Modification)
In the above embodiment, an object that is obstructed by other objects by calculating the number of objects hidden by the image of other objects using the rendering unit, and performing a search using the number of objects as the upper limit. Was detected. As another method, the following method may be used. The search result number estimation unit (calculation unit) uses the rendering unit to identify an object hidden by an image of another object of at least a part of the image. The object search unit inspects only for the identified object whether the object is blocked by another object. When the number of objects to be identified is small, an efficient search can be performed also by this method. In the example of FIGS. 4 and 5, the search result number estimation unit identifies the object f on the assumption that some images are hidden by the images of the other objects e. Then, the object search unit determines whether the object f is actually blocked by other objects in the space. Here, it is determined that the object f is blocked by the object e.

  As described above, according to the present embodiment, it is possible to efficiently search for objects that are blocked by other objects. For example, a PV (Photovoltaic power generation) device or a building can be used as an object, and a PV device behind a building such as a building can be searched. As a result, it is possible to examine the PV arrangement in consideration of whether it is a shadow. In addition, high-speed 3D rendering becomes possible. It can also be applied to CAD in general, such as machine component placement and building design.

  In addition, the object search apparatus in this embodiment demonstrated above is realizable by using a general purpose computer apparatus as basic hardware, for example. That is, each processing unit of the object search device can be realized by causing a processor mounted on the computer device to execute a program. At this time, the object search device may be realized by installing the above program in a computer device in advance, or may be stored in a storage medium such as a CD-ROM or distributed through the network. Thus, this program may be realized by appropriately installing it in a computer device. The storage unit 102 in the object search device may be configured by an internal or external memory device and a hard disk, or a recording medium such as a CD-R, CD-RW, DVD-RAM, or DVD-R. .

Claims (8)

A database for storing arrangement information of a plurality of objects in space;
A rendering unit that draws the plurality of objects from a viewpoint set in the space, and a calculation that performs rendering using the rendering unit and calculates the number of objects in which at least some images are hidden by images of other objects And
Based on the database, a search operation is performed based on the database, using the number calculated by the calculation unit as an upper limit value of the number of search results, and searching for an object in the space that is at least partially obstructed by another object when viewed from the viewpoint. An object search unit to perform,
An object search device comprising: 2. The object search device according to claim 1, wherein the object search unit ends the search calculation when the number of objects of the upper limit value is searched. 3. The object search device according to claim 1, wherein the object search unit selects a search strategy from a plurality of search strategies based on the upper limit value, and executes a search operation according to the selected search strategy. The calculation unit uses the rendering unit,
Generating first drawing data in which all of the plurality of objects are drawn;
Generating second drawing data each independently drawing one of the plurality of objects,
When an image of an individual object indicated in the first drawing data does not match an image of the second drawing data corresponding to the individual object, the image of the individual object is hidden by an image of another object. The object search device according to claim 1, wherein the object search device is determined to be. The calculation unit uses the rendering unit,
The plurality of objects are drawn in order for each pixel, and when another object is already drawn on the pixel to be drawn, by determining the context of the drawn object and the other object, 2. The object search device according to claim 1, wherein it is determined which of the drawn object and the another object is hidden. The storage unit assigns a different one of the objects to each leaf node, and stores a hierarchical object structure in which an envelope containing an object of each child node of the parent node is assigned to a parent node,
The object search unit, as the first search strategy,
(A) determining whether there is a possibility that an object obstructed by another object is included in an outer package corresponding to each of the child nodes of the parent node;
(B) Repeating the same determination as in (A) above for child nodes that are determined to exist,
(C) For the arrived leaf node, determine whether the object corresponding to the leaf node is blocked by another object,
6. The object search device according to any one of claims 1 to 5. The storage unit assigns a different one of the objects to each leaf node, and stores a hierarchical object structure in which an envelope containing an object of each child node is assigned to a parent node,
The object search unit as a second search strategy,
(A) Select one of the child nodes of the parent node,
(B) Judge whether there is a possibility that an object obstructed by other objects is included in the outer envelope corresponding to the selected node.
(C) When it is determined that there is no possibility of being included, the other node is selected from the child nodes, the determination of (B) is performed for the selected node,
(D) When it is determined that there is a possibility of being included in (B) or (C), the processing after (A) is recursively performed on the child node of the selected node, and the arrived leaf For a node, determine if the leaf node object is blocked by another object,
5. The object search device according to any one of claims 1 to 4. Accessing a database storing arrangement information of a plurality of objects in space;
A rendering step of drawing the plurality of objects from a viewpoint set in the space;
A calculation step of calculating the number of objects in which at least a part of the image is hidden by an image of another object based on the rendering by the rendering step;
Based on the database, a search operation for searching for an object that is at least partially obstructed by another object as viewed from the viewpoint in the space is performed using the number calculated in the calculating step as an upper limit value of the number of search results. Object search step to perform,
Object search method comprising:

Images