JP2017166971A - Object detection device and object detection program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an object detection device and an object detection program designed to improve the accuracy of detecting even a black object of low reflectivity when such an object exists on a road.SOLUTION: A control device of the present invention acquires distance information in front of the own vehicle by a three-dimensional distance detection unit such as a laser radar (S2), and maps it to a bird's eye view grid (S3). The control device detects a cell in which a solid object exists from point cloud information (S4, S5), and detects a road surface with respect to a cell at a position closer than is the cell (S6, S7). The control device calculates an estimated solid object by an expression (A) with respect to an unspotted cell (S8, S9), and sets an estimated weight (S10). The control device converts the bird's eye view grid to a static grid (S11) and applies similar processing after the own vehicle moves, and thereby adds solid object probability (S12). In this way, it is possible to determine for even an unspotted cell that a solid object exists, providing that there is an increase in the solid object probability.SELECTED DRAWING: Figure 5

Description

  The present invention relates to an object detection device and an object detection program.

  2. Description of the Related Art There is an object detection device that detects a three-dimensional object existing ahead when a vehicle travels on a road by using LIDAR (light detection and ranging, laser imaging detection and ranging; laser radar). For example, a laser radar is provided in the front part of a passenger compartment of an automobile, and the distance is measured by the reflected light by irradiating the road surface ahead.

  Detect the distance to the measurement point from the host vehicle in various directions ahead, and if there are multiple measurement points with the same distance, detect the presence of the three-dimensional object ahead by calculating the height dimension. be able to. In this case, the presence of the three-dimensional object is detected by the distance information based on the reflected light, but if the distance information to the road surface of the projection destination is obtained, the presence of the three-dimensional object is detected between the vehicle and the three-dimensional object. Since it is not done, it can be determined that the road surface.

  However, if there is actually a black object with low reflectivity on the road or an object that does not return reflected light to the light projection side, the area where distance information is not obtained or the number of sufficient distance information is not obtained Is assumed to occur. In this case, even if a three-dimensional object is present, there is a case where it cannot be detected, and since the three-dimensional object is not detected, there is a possibility that the road surface is erroneously determined.

JP 2013-140515 A

  The present invention has been made in view of the above circumstances, and its purpose is to estimate whether or not this region includes a three-dimensional object when there is a region where distance information by reflected light cannot be obtained. It is an object to provide an object detection apparatus and an object detection program that can be used.

  The object detection apparatus according to claim 1 is divided by a three-dimensional distance detection unit (3) that acquires at least three-dimensional distance information ahead of the host vehicle and a bird's-eye grid based on the viewpoint of the three-dimensional distance detection unit. A three-dimensional object detection unit (4) that detects the presence or absence of a three-dimensional object based on the three-dimensional distance information for each cell, and distance information between the host vehicle and the three-dimensional object when the three-dimensional object exists. When there is a pointless area including a pointless cell with no distance information between the host vehicle and the three-dimensional object, there is a pointless region. A three-dimensional object estimation unit (6) for estimating an estimated three-dimensional object having a maximum height T1 that can exist in the interior by the following equation (A), and converting at least the cell of the estimated three-dimensional object into a stationary grid cell based on a road surface And the three-dimensional distance detection unit as the vehicle moves When the estimated three-dimensional cell estimated by the three-dimensional object estimation unit is converted from the new three-dimensional distance information obtained to the stationary grid, it is detected as a three-dimensional object when repeatedly estimated at the same road surface position. And a three-dimensional object determination unit (7).

T1 = T0 · (re−r1) / re (A)
Where T0 is the height of the three-dimensional distance detection unit, re is the far point distance of the astigmatic area, and r1 is the near point distance of the astigmatic area.

  By adopting the above configuration, the three-dimensional distance detection unit acquires at least the three-dimensional distance information ahead of the host vehicle, and applies the acquired three-dimensional distance information to each cell divided by the bird's-eye view grid. The presence or absence of a three-dimensional object is detected for each cell. When the presence of the three-dimensional object is detected, the road surface detection unit detects a pointed cell having distance information among the cells between the host vehicle and the three-dimensional object as the road surface. In addition, when there is a pointless area including a pointless cell with no distance information between the host vehicle and the three-dimensional object, the three-dimensional object estimation unit is the maximum that can exist in the pointless area according to the formula (A). An estimated solid object having a height T1 is estimated. The three-dimensional object determination unit converts at least the estimated three-dimensional object cell into a stationary grid cell, and is estimated by the three-dimensional object estimation unit from new three-dimensional distance information obtained by the three-dimensional distance detection unit as the host vehicle moves. When the estimated solid object cell is converted to a stationary grid and repeatedly estimated at the same road surface position, it can be detected as a solid object. As a result, even when there is a pointless region where the three-dimensional distance information is not obtained by the three-dimensional distance detection unit, it is possible to estimate whether or not a three-dimensional object exists in the pointless region, and an erroneous determination occurs. Can be reduced.

Functional block diagram showing the first embodiment (A) Elevation view and (b) Plan view showing detection range and bird's-eye view grid of three-dimensional distance detection unit The figure which shows the example which showed the pointed distance information on the bird's-eye view grid Stationary grid layout Flow chart of object detection processing Illustration of detecting the height of a solid object in the bird's-eye view grid Action explanatory diagram to estimate the maximum height of the pointless range of the bird's eye grid Examples of presumed solid objects with different pointless areas Illustration of the width of the astigmatism area and the height of the estimated solid object Partial view showing a specific example of distribution of pointed distance information of bird's-eye view grid Partial view showing a specific example of the score determination result of the bird's-eye view grid (A) Explanatory drawing when the bird's eye grid is superimposed on the stationary grid, (b) Partial view of the bird's eye grid Score distribution map of stationary grid (A) Operation explanatory diagram when the bird's-eye grid after movement is superimposed on the stationary grid, (b) Partial view of the bird's-eye grid Point distribution chart when adding points after moving to a stationary grid The figure which shows the specific example of the score determination result of the bird's-eye view grid which shows 2nd Embodiment. Action explanatory diagram to estimate the maximum height of each pointless cell of the bird's eye view grid (A) Explanatory drawing when the bird's eye grid is superimposed on the stationary grid, (b) Partial view of the bird's eye grid Score distribution map of stationary grid (A) Operation explanatory diagram when the bird's-eye grid after movement is superimposed on the stationary grid, (b) Partial view of the bird's-eye grid Point distribution chart when adding points after moving to a stationary grid The figure which shows the specific example of the score determination result of the bird's-eye view grid which shows 3rd Embodiment. (A) Explanatory drawing when the bird's eye grid is superimposed on the stationary grid, (b) Partial view of the bird's eye grid Score distribution map of stationary grid (A) Operation explanatory diagram when the bird's-eye grid after movement is superimposed on the stationary grid, (b) Partial view of the bird's-eye grid Point distribution chart when adding points after moving to a stationary grid

(First embodiment)
Hereinafter, a first embodiment of the present invention will be described with reference to FIGS.
2 (a) and 2 (b), the object detection device 1 is mounted on, for example, the upper front portion of the vehicle A, and is provided on the inner upper portion of the windshield in the vehicle interior. As shown in FIG. 1, the object detection device 1 includes a three-dimensional distance detection unit 3, a storage unit 4, and an output unit 5 with a control device 2 as a main component.

  The control device 2 includes a CPU, a ROM, a RAM, an interface, and the like, and may be incorporated in an in-vehicle ECU or may be provided separately. The control device 2 detects a three-dimensional object existing in front of the vehicle A from the distance information detected by the three-dimensional distance detection unit 3 based on an object detection program described later. In FIG. 1, the structure by the function of the control apparatus 2 is shown. The constituent elements include a solid detection unit 2a, a road surface detection unit 2b, a solid object estimation unit 2c, and a solid object determination unit 2d.

  The three-dimensional distance detection unit 3 is a so-called laser radar (LIDAR). For example, a distance measuring laser beam is irradiated in various directions in front of the vehicle A to set the distance of the traveling range in front 3. Dimensionally detected. The memory | storage part 4 can also be provided in the control apparatus 2, and can also be set as the structure separately provided outside. The storage unit 4 stores an object detection program executed by the control device 2 and various information. The output unit 5 displays the detection result of the three-dimensional object, and can be displayed using a display or the like provided in the vehicle interior.

  The three-dimensional distance detection unit 3 irradiates the front road surface R of the vehicle A with laser light as shown in FIG. Reflected light while moving the irradiation point P from the near point side to the far point side at substantially equal intervals, such as the distances r1, r2,..., Rn on the road surface R from the near point side with the position immediately below the front surface of the vehicle A as the origin (O). The distance is calculated by detecting the arrival time. Further, as shown in FIG. 2B, the three-dimensional distance detection unit 3 measures the distance in the fan-shaped detection area by changing the direction to the left and right with the front r direction as the center and oscillating the laser beam. .

  In this case, when the road surface R is irradiated with laser light, information on the distance corresponding to the irradiation point P on the road surface R is obtained. On the other hand, when an object such as a three-dimensional object exists on the road surface R, the laser light is irradiated onto the three-dimensional object before reaching the road surface R, and the reflected light is received to calculate the distance. The distance information of the position closer than the irradiation point P is obtained.

  As shown in FIG. 2B, the distance information obtained by the three-dimensional distance detection unit 3 includes a plurality of cells Gθn (divided by the bird's eye grid G corresponding to the vehicle A according to the irradiation range of the laser light. For example, the three-dimensional object is determined in the direction θ: a to e and the distance n: 0 to 4). The bird's-eye view grid G divides a range of a predetermined angle including the r-axis with respect to the r-axis indicating the traveling direction of the vehicle A as cells Ga0 to Ga4 at predetermined distances from the side close to the vehicle A. Similarly, the cells Gb0 to Gb4 are set in a predetermined angular range on the left side of the cells Ga0 to Ga4 with respect to the traveling direction of the vehicle A, and the cells Gc0 to Gc4 are set on the left side. Further, the cells Gd0 to Gd4 are set in a predetermined angular range on the right side of the cells Ga0 to Ga4 with respect to the traveling direction of the vehicle A, and the cells Ge0 to Ge4 are set on the right side thereof.

  The distance information obtained by the reflection of the laser light is associated with the cell Gθn of the bird's eye grid G according to the distance, as shown in FIG. In the figure, positions indicating distance information in corresponding cells are indicated by black circles (dots).

  On the other hand, as shown in FIG. 4, for the determination of the three-dimensional object, a stationary grid S of an orthogonal coordinate system corresponding to the road surface R is set. The stationary grid S is a rectangular matrix cell Sxy (for example, from S00 to S77) corresponding to the position in the x direction and y direction, where x is the direction orthogonal to the traveling direction of the vehicle on the road surface R and y is the traveling direction. 8 × 8 cells) are set. In FIG. 4, the position of the object detection device 1 of the vehicle A is shown on the stationary grid S.

  Next, object detection processing by the object detection apparatus 1 will be described with reference to FIG. The flowchart of this object detection process is stored in the storage unit 4 as an object detection program. The control device 2 reads the object detection program from the storage unit 4 when starting the object detection process.

  When starting the program, the control device 2 first initializes data on the stationary grid S in step S1, and sets the stationary grid S in a region including the road surface A ahead corresponding to the position of the host vehicle A. Next, when the control device 2 proceeds to step S2, the laser radar of the three-dimensional distance detection unit 3 irradiates laser light toward the front of the vehicle A (r direction in the bird's eye grid G; y direction in the stationary grid S). And distance information is acquired corresponding to the position of the bird's-eye view grid G. In this case, the control device 2 acquires point group information of distance information obtained by the three-dimensional distance detection unit 3. The point group information is information on the irradiation angle and distance of the laser light, and is information on polar coordinates centered on the position of the three-dimensional distance detection unit 3.

  The control device 2 maps the acquired point cloud information on the bird's-eye view grid G in step S3. At this time, when the distance information is shorter than the distance on the road surface R, the information is mapped to the cell Gθn according to the distance. For example, as shown in FIG. 6, if the distance of a point Pb corresponding to a position farther than the point Pa near the road surface R is the same, a solid object that blocks and reflects the laser beam here. It can be estimated that it exists.

  When the height Ha at this time is equal to or greater than a predetermined threshold set as a threshold, for example, 5 to 10 cm or more, there is a possibility that the vehicle A may become an obstacle to ride on when the vehicle A travels. In such a case, it can be detected that there is a three-dimensional object. Thus, the distance information point P belonging to the same cell Gθn is associated with each cell Gθn of the bird's-eye view grid G as shown in FIG.

  Next, the control apparatus 2 performs a three-dimensional object detection process at step S4. Here, as described with reference to FIG. 6, when there are a plurality of distance information and the height H is calculated for the distance information that falls within one cell Gθn, and there is a height equal to or higher than the threshold value. The presence of a three-dimensional object is detected. Specifically, the control device 2 obtains the height of the measurement point based on the distance information in the cell Gθn for a plurality of distance information in the cell Gθn.

  Here, the control device 2 calculates the difference between the lowest height point Pa and the highest height Pb from the height information of the plurality of measurement points to obtain the height Ha. The control device 2 determines that a three-dimensional object exists in the cell Gθn when the obtained value of the height Ha is equal to or greater than a threshold value Ho for determining the three-dimensional object. Subsequently, the control device 2 proceeds to step S5, and sets “+1” as the score of the three-dimensional object probability for the cell Gθn in which the three-dimensional object is detected.

  Next, the control apparatus 2 performs a road surface detection process at step S6. Specifically, the control device 2 determines the road surface R of the cells Gθ0 to Gθ (n−1) between the host vehicle A and the cell Gθn where the three-dimensional object is detected for which distance information is obtained. Is determined to be a detected cell. Subsequently, in step S7, the control device 2 sets “−1” as the solid object probability score for the cell determined as the road surface among the cells Gθ0 to Gθ (n−1).

  Next, in step S8, the control device 2 performs the following process on the cell Gθk (k: integer of 0 ≦ k <n) for which distance information is not obtained. That is, for the direction θ of the cell Gθn in which the presence of the three-dimensional object is detected, the control device 2 obtains the cell that is not determined as the road surface among the cells Gθ0 to Gθ (n−1) before the three-dimensional object, that is, the distance information. A cell Gθk (k: integer of 0 ≦ k <n) is recognized as an astigmatism region NP. In this case, the astigmatism region NP may be a single cell or a plurality of continuous cells. In the astigmatism region NP, there may be a solid object that does not reflect the laser light or has a low reflectance in the region.

Therefore, the control device 2 first performs an estimation process for the highest one when a three-dimensional object is present in the astigmatism region NP. The calculation principle will be described with reference to FIG. When the far point distance of the astigmatic area NP is re, the near point distance is r1, and the height of the object detection device 1 is T0, the highest three-dimensional object that exists in the astigmatic area NP exists at the near point position r1. Therefore, the height T1 can be calculated by the following equation (A).
T1 = T0 · (re−r1) / re (A)
Where T0 is the height of the three-dimensional distance detector, re is the far point distance of the astigmatic area NP, and r1 is the near point distance of the astigmatic area NP.

  In this case, what can be estimated by the above formula (A) is the height of the highest three-dimensional object that can exist, and in fact, a three-dimensional object with a height less than that may exist. For example, as shown in FIG. 8, when there are three continuous cells Ga3 to Ga5 as the astigmatic region NP, the three-dimensional object Xa having the highest Ha height exists at the position near the cell Ga3 ( a) may occur. In some cases (b), there is a three-dimensional object Xb having a low height Hb so as to straddle the cells Ga3 to Ga5. Furthermore, there is a case (c) where there is a three-dimensional object Xc having a height Hc extending over the two cells Ga3 and Ga4.

  On the other hand, as shown in FIG. 9, when a three-dimensional object that can exist has a short depth, the number of continuous dotless cells is determined according to the height. For example, in a three-dimensional object having a height Ha as shown in FIG. 9A, three continuous pointless cells Ga3 to Ga5 are generated. Further, as shown in FIG. 9B, in the three-dimensional object having the height Hd, one pointless cell Ga5 is generated. When there is one dotless cell, the maximum height may be smaller than a threshold value depending on the depth dimension of the cell, and it may not be determined as a three-dimensional object.

  From the above, the control device 2 estimates that a three-dimensional object is present in this astigmatic area when the maximum height of the astigmatic area exceeds the threshold in step S9, and sets it as an estimated three-dimensional object. Subsequently, in step S10, the control device 2 sets an estimated weight α for the cell in which the presence of the estimated three-dimensional object is estimated. Here, for example, a value determined in the range of 0 to 1 is set as the solid object probability as the estimated weight α of the estimated solid object.

  Next, the control device 2 proceeds to step S11 and performs a process of converting the result obtained for each cell of the bird's eye grid G of the polar coordinate system into a cell of the stationary grid S of the orthogonal coordinate system. In this case, it is possible to perform processing for associating with the stationary grid S corresponding to the position of the road surface R directly, and assuming a moving bird's-eye view grid in an orthogonal coordinate system with the position of the object detection device 1 of the host vehicle A as the origin. Conversion processing can also be performed.

  In the case of conversion to a bird's-eye view grid that moves in an orthogonal coordinate system, the same process can be performed as a result by performing the process of applying the converted orthogonal coordinate system to the stationary grid S corresponding to the road surface R. Next, the control device 2 adds the value of the three-dimensional object probability obtained as described above to each cell Sxy on the stationary grid S.

  The control device 2 proceeds to step S13, and changes and sets the position of the host vehicle A in the stationary grid S based on the position information of the host vehicle A. Thereafter, the control device 2 returns to step S2, performs the same process as the above process based on the point cloud information of the position after the movement, and thereafter repeatedly executes steps S2 to S13.

  The control device 2 performs a three-dimensional object determination process separately from the above process. As a result of repeating the above processing, the solid object probability is added to the cell Sxy of the stationary grid S, so that there is a solid object that is stationary on the road surface R or there is an estimated solid object. The value of the three-dimensional object probability of the cell increases. Thereby, the control apparatus 2 can estimate the stationary solid thing which exists in the road surface R correctly.

  In addition, when there is a moving three-dimensional object such as another vehicle, the three-dimensional object moves with the movement of the other vehicle, so that the value of the three-dimensional object probability does not increase but moves between cells. Become. Furthermore, if there is no solid object in the astigmatism region, the value of the solid object probability does not increase by a plurality of addition processes, so the presence of the three-dimensional object is not determined.

Next, specific examples will be described with reference to FIGS.
FIG. 10 shows eight cells Ga <b> 0 to Ga <b> 7 corresponding to the r direction (the column direction of the cells Gan) in the bird's eye view grid G. As described above, the control device 2 takes in the distance information by the laser beam with respect to the measurement point P acquired by the three-dimensional distance detection unit 3 in step S2 through step S1 of the program shown in FIG. It is out. Actually, as shown in FIG. 3, the point cloud information is obtained for all the cells of the bird's-eye view grid G, and the control device 2 executes the following processing for all of these cells. The process will be described with respect to the eight cells Ga0 to Ga7 shown in FIG.

  The control device 2 maps the point cloud information to all the cells of the bird's-eye view grid G in step S3. FIG. 10 is an example of mapping in eight cells in the r direction. Here, there are pointed cells Ga0, Ga1, Ga5, Ga6 from which distance information is obtained, and pointless cells Ga2-Ga4, Ga7 from which distance information is not obtained.

  The control device 2 performs the three-dimensional object detection process in step S4 for the pointed cell Ga6 including distance information shorter than the distance to the irradiation point P on the road surface R among the pointed cells Ga0, Ga1, Ga5, Ga6. Go and detect solid objects. In step S5, the control device 2 sets a three-dimensional object probability “+1” for the three-dimensional object cell Ga5 including the three-dimensional object. As shown in FIG. 11, this result is mapped so as to set “+1” to the three-dimensional object cell Ga6. The control device 2 performs the same three-dimensional object processing for the other pointed cells Ga0, Ga1, and Ga5. However, when the distance information is approximately equal to the distance to the irradiation point P on the road surface R, the three-dimensional object is not detected.

  Next, since the control device 2 is located between the three-dimensional object cell Ga5 and the host vehicle A, the pointed cells Ga0, Ga1, and Ga5 that are not determined to be three-dimensional object cells are detected as road surface cells in step S6. The controller 2 sets “−1” as the three-dimensional object probability in step S7 for the road surface cells Ga0, Ga1, and Ga5. As shown in FIG. 11, this result is mapped so as to set “−1” to the road surface cells Ga0, Ga1, and Ga5.

  Subsequently, the control device 2 performs a process of estimating the highest height T1 when there is a three-dimensional object in step S8 for the astigmatism region NP including the astigmatism cells Ga2 to Ga4. In this case, the far point distance re of the pointless region NP of the formula (A) is a distance from the position of the object detection device 1 to the far point of the pointless cell Ga4. The near point distance r1 of the astigmatism region NP is a distance to the near point of the astigmatism cell Ga2. When the calculated height T1 exceeds the threshold value, this is set as an estimated three-dimensional object.

  When it is determined that the estimated solid object is present in the astigmatic area NP, the control device 2 sets the estimated weight of the estimated solid object in each of the cells Ga2 to Ga4 of the astigmatic area NP. In this setting process, the control device 2 multiplies the solid object probability “+1” of the solid object by, for example, “0.5” as the estimated weight α. As a result, the control device 2 sets the three-dimensional object probability “+0.5” of the estimated three-dimensional object to the pointless cells Ga2 to Ga4.

  By performing the above processing on the entire bird's-eye view grid G, information on a three-dimensional object, a road surface, and an estimated three-dimensional object existing in front of the host vehicle A can be obtained. In FIG. 11, the result about eight cells Ga0-Ga7 among the bird's-eye view grids G is shown. Note that, for the pointless cell Ga7 in which distance information ahead of the cell Ga6 in which the three-dimensional object is detected is not obtained, the existence of the three-dimensional object is unknown, so “0” is set as the three-dimensional object probability.

  12 to 15 show specific examples in the case of converting information detected in the bird's-eye view grid G into a stationary grid S. Here, for the sake of simplicity, as shown in FIG. 12B, in the bird's eye grid G, a three-dimensional object is detected in the pointed cell Ga4, and among the cells Ga0 to Ga3 located in front of the pointed cell Ga0, Ga3, Is described as a case where the estimated solid object is estimated in the pointless cells Ga1 and Ga2.

  In the bird's-eye view grid G shown in FIG. 12B, the solid object probability is “+1” in the pointed cell Ga4 in which the three-dimensional object is detected, “−1” in the pointed cells Ga0 and Ga3 in which the road surface is detected, and the estimated three-dimensional object. "+0.5" is set in the pointless cells Ga1 and Ga2 in which is detected. For other cells, “+1” is set in the cell in which the three-dimensional object is detected, and “−1” is set in the cell in which the road surface is detected.

  The control device 2 converts the cell information of the bird's eye grid G into cell information of the stationary grid S in step S11. Each cell Sxy of the stationary grid S is set to “0” in the initial state. In addition, the position of the object detection device 1 that is the origin position of the bird's-eye view grid G is at the position of the cell S30. As shown in FIG. 12A, when the bird's-eye view grid G is superimposed on the stationary grid S, for example, the astigmatism cell Ga1 corresponds to the cell S31, and the astigmatism cell Ga2 corresponds to the cell S32. In step S12, the control device 2 adds the converted three-dimensional object probability data to the stationary grid S to obtain data as shown in FIG.

  Thereafter, the control device 2 performs processing so as to shift the vehicle position on the stationary grid S based on the current position information of the vehicle A. Thereby, for example, as shown in FIG. 14A, it is recognized that the position of the object detection device 1 of the host vehicle A has moved to the cell S42 on the stationary grid S and is shifted.

  Hereinafter, the point cloud information of the bird's-eye grid G at the position of the movement destination is obtained by repeatedly executing the same processing as described above, and the control device 2 uses the point cloud information to generate the point cloud of the new bird's-eye grid G. get information. FIG. 14B shows the distribution state of the dotless cells at the movement destination. Here, the pointless cells Gb1, Gc0 to Gc2 in which the estimated three-dimensional object is estimated are detected, and the three-dimensional object probability “+0.5” is set for each of them.

  When such cell information of the bird's-eye grid G is converted in association with the cell information of the stationary grid S, the result is as shown in FIG. Here, a solid object probability “+0.5” is set for the cells S32, S23, and S33 of the stationary grid S for the four pointless cells Gb1 and Gc0 to Gc2.

  Next, the control device 2 adds the solid object probability data of the stationary grid S shown in FIG. 14 obtained as a result of the above to the stationary grid S shown in FIG. As a result, the added stationary grid S as shown in FIG. 15 is obtained. As can be seen from FIG. 15, by adding the solid object probability twice, the cell S31 of the stationary grid S is “+0.5”, the cell S32 is “+1”, the cell S23 is “+1.5”, and the cell S33 is “−0.5”.

As a result, the three-dimensional object probability of the cells S32 and S23 is high, and the probability that the estimated three-dimensional object is a three-dimensional object is high. Even if it is a pointless cell for which distance information is not obtained, the three-dimensional object The presence of can be detected.
In the determination from the estimated three-dimensional object to the three-dimensional object, the point cloud information of the bird's-eye grid G can be obtained twice and obtained as described above, or can be obtained from three or more acquisition results. it can.

  According to the first embodiment, when the control device 2 determines a three-dimensional object on the road surface R based on the distance information detected by the three-dimensional distance detection unit 3, no distance information is obtained. When there is a cell, assuming the presence of a three-dimensional object, the existence of the highest estimated three-dimensional object is estimated, and the three-dimensional object probability is set by multiplying by the estimated weight α (= 0.5). As a result, by adding a plurality of results on the stationary grid S, it is possible to determine that a three-dimensional object exists for a cell whose solid object probability has increased.

  As a result, even when there is a pointless cell around which the distance information by the laser light irradiated by the laser radar is not obtained around the host vehicle A, it is possible to estimate the presence of the three-dimensional object on the road surface R. The estimation of the existing three-dimensional object can be performed appropriately.

(Second Embodiment)
FIGS. 16 to 21 show the second embodiment, and only the parts different from the first embodiment will be described below. In this embodiment, the detection accuracy can be increased in the processing method in the case where there are undotted cells.

  FIG. 16 corresponds to FIG. 11 shown in the first embodiment. When point cloud information is obtained in the same manner as in FIG. 10, in this embodiment, the control device 2 uses the object shown in FIG. Processing related to the estimated three-dimensional object in steps S9 and S10 of the detection processing is performed as follows. That is, in the calculation process of the estimated three-dimensional object in the astigmatic area NP, the control device 2 estimates a three-dimensional object for each astigmatism cell when the astigmatism cells are continuous.

As illustrated in FIG. 17, for example, when the astigmatic region NP includes n continuous astigmatic cells, the control device 2 estimates the highest three-dimensional object for each of the n astigmatic cells. Here, as in the first embodiment, the far point distance of the astigmatism region NP is set to re. The near point position of the nearest dotless cell in the dotless region NP is r1, and the near point position of the kth (n ≧ k> 1) dot cell is rk. As described above, assuming that the height of the three-dimensional distance detection unit 3 is T0, assuming that the highest three-dimensional object that can exist in each pointless cell exists at the near point position rk, the height Tk is It can be calculated by the following formula (B).
Tk = T0 · (re−rk) / re (B)
In this case, the formula (B) is the same formula as the formula (A) for the closest pointless cell. In addition, for a pointless cell at a far position, the difference between the numerator value, that is, the distance from the far point re becomes small, and thus the height Tk of the estimated three-dimensional object may be below the threshold value.

  Therefore, for example, in the example shown in FIG. 16, among the three pointless cells Ga2 to Ga4 of the pointless region NP, the pointless cells Ga2 and Ga3 on the closer side are estimated solid objects that are equal to or higher than the threshold value. “0.5” is set as the three-dimensional object probability. On the other hand, the astigmatism cell Ga4 at a position far from the astigmatism region NP is an estimated three-dimensional object lower than the threshold value, and thus is regarded as a road surface, and “−1” is set as the three-dimensional object probability.

  18 to 21 show specific examples in the case where information detected in the bird's-eye view grid G is converted into a stationary grid S, as in the first embodiment. FIG. 18B is a distribution diagram of the point cloud information acquired under the same conditions as in FIG. 12B. In this embodiment, the two dot-free cells Ga1 and Ga2 in the dot-free region NP are shown. Of these, a case will be described in which the estimated solid object is estimated in the pointless cell Ga1 and the road surface is estimated in the pointless cell Ga2 by the control device 2 in step S9.

  In the bird's-eye view grid G shown in FIG. 18B, among the two pointless cells Ga1 and Ga2 in the pointless region NP, “+0.5” is set in the pointless cell Ga1 in which the estimated three-dimensional object is detected, "-1" is set in the pointless cell Ga2 for which the road surface is determined.

  When the control device 2 converts the cell information of the bird's eye grid G into the cell information of the stationary grid S in step S11 and then superimposes the bird's eye grid G on the stationary grid S, the control device 2 becomes as shown in FIG. As a result, as shown in FIG. 19, the three-dimensional object probability “+0.5” of the pointless cell Ga1 is converted into the cell S31, and the three-dimensional object probability “−1” of the pointless cell Ga2 is converted into the cell S32.

  After that, the control device 2 performs the same processing after shifting the position of the host vehicle A, so that the point cloud information of the new bird's-eye grid G at the destination position is shown in FIG. The distribution state of such pointless cells is obtained. As a result of calculating the calculation processing of the estimated three-dimensional object according to the formula (B), the pointless cells Gb1 and Gc2 are determined as road surfaces, so “−1” is set, and the pointless cells Gc0 and Gc1 are determined to be estimated three-dimensional objects. “+0.5” is set.

  When such cell information of the bird's-eye grid G is converted in association with the cell information of the stationary grid S, the result is as shown in FIG. Here, among the four pointless cells Gb1 and Gc0 to Gc2, the solid object probability “+0.5” for the cell S32 of the stationary grid S with respect to the pointless cells Gc0 and Gc1 in which the estimated three-dimensional object is set. Is set.

  Next, the control device 2 adds the data of the three-dimensional object probability of the stationary grid S shown in FIG. 20 obtained as the above result to the stationary grid S shown in FIG. As a result, the added stationary grid S as shown in FIG. 21 is obtained. As can be seen from FIG. 21, by adding the solid object probabilities twice, the cell S31 of the stationary grid S becomes “+0.5”, the cell S32 is “−0.5”, and the cell S23 is “0”. The cell S33 becomes “−2”.

  As a result, since the solid object probability of the cell S32 is “+0.5”, the existence of the solid object cannot be confirmed, but the possibility remains. In addition, since the other unspotted cells S32, S23, and S33 have a negative solid object probability, the possibility of being a road surface is high.

  According to the second embodiment, it is possible to obtain the same effect as that of the first embodiment. In addition, when the control device 2 makes the astigmatism region NP composed of a plurality of astigmatism cells, the Since the highest estimated three-dimensional object is calculated for each cell, it becomes possible to make a finer determination for each cell.

(Third embodiment)
22 to 26 show the third embodiment, and only the parts different from the second embodiment will be described below. In this embodiment, the method of setting the three-dimensional object probability is set in accordance with the height of the estimated three-dimensional object of the pointless cell.

  FIG. 22 corresponds to FIG. 16 shown in the second embodiment, and when point cloud information is obtained in the same manner as in FIG. 10, in this embodiment, the control device 2 uses the object shown in FIG. Processing related to the estimated three-dimensional object in steps S9 and S10 of the detection processing is performed as follows. That is, in the calculation process of the estimated three-dimensional object in the astigmatic area NP, the control device 2 estimates a three-dimensional object for each astigmatism cell when the astigmatism cells are continuous.

  Here, the control device 2 calculates the highest value of the estimated three-dimensional object based on the formula (B) in the same manner as in the second embodiment. As a result, in the astigmatism region NP, as shown in FIG. 17, the height of the astigmatism cell closer to the object detection device 1 is higher, and becomes lower as the distance increases. In accordance with the height of the estimated three-dimensional object, for example, the control device 2 sets the estimated weight α as large as “0.8” if it is high, and sets the estimated weight α as “0.2” if it is low. It is set low. As a result, in step S10, the control device 2 sets the three-dimensional object probability corresponding to each cell as shown in FIG.

  23 to 26 show specific examples in the case where information detected in the bird's-eye grid G is converted into a stationary grid S, as in the second embodiment. FIG. 23B is a distribution diagram of point cloud information acquired under the same conditions as in FIG. 18B, but in this embodiment, the control device 2 performs the process in step S9 in the astigmatic region NP. The estimated three-dimensional object is estimated by the two pointless cells Ga1 and Ga2, and the three-dimensional object probability is shown in which the estimated weight α is associated with the estimated three-dimensional object depending on the height. Here, for example, a case where “+0.5” is set in the pointless cell Ga1 and “+0.3” is set in the pointless cell Ga2 will be described.

  When the control device 2 converts the cell information of the bird's-eye grid G into the cell information of the stationary grid S in step S11 and then superimposes the bird's-eye grid G on the stationary grid S, the control device 2 becomes as shown in FIG. As a result, as shown in FIG. 24, the three-dimensional object probability “+0.5” of the pointless cell Ga1 is converted into the cell S31, and the three-dimensional object probability “+0.3” of the pointless cell Ga2 is converted into the cell S32.

  Thereafter, the control device 2 performs the same processing after shifting the position of the host vehicle A, and the point cloud information of the new bird's-eye grid G at the destination position is shown in FIG. The distribution state of such pointless cells is obtained. As a result of calculating the calculation process of the estimated three-dimensional object according to the equation (B), the estimated three-dimensional object is detected in all of the pointless cells Gb1, Gc0 to Gc2. That is, “+0.2” is set for the pointless cell Gb1, “+0.5” for the pointless cell Gb0, “+0.2” for the pointless cell Gc1, and “+0.1” for the pointless cell Gc2.

  When the cell information of the bird's-eye grid G is converted in association with the cell information of the stationary grid S, the result is as shown in FIG. Here, for the four pointless cells Gb1 and Gc0 to Gc2, the three-dimensional object probability is “+0.5” for the cell S32 of the stationary grid S, “+0.2” for the cell S33, and the cell “+0.1” is set for S23.

  Next, the control device 2 adds the solid object probability data of the stationary grid S shown in FIG. 25A obtained as a result of the above to the stationary grid S shown in FIG. As a result, an added stationary grid S as shown in FIG. 26 is obtained. As can be seen from FIG. 26, by adding the solid object probability twice, the cell S31 of the stationary grid S is “+0.5”, the cell S32 is “+0.8”, the cell S23 is “+1.1”, the cell S33 becomes “−0.8”.

  As a result, the three-dimensional object probability is high in the cells S32 and S23, and the possibility that a three-dimensional object exists is high. In addition, the other pointless cell S33 has a negative three-dimensional object probability, and thus has a high possibility of being a road surface.

  According to such 3rd Embodiment, while being able to obtain the same effect as 2nd Embodiment, the control apparatus 2 of the estimated solid object from which the astigmatic area | region NP is calculated from a several astigmatism cell Since the three-dimensional object probability is obtained by multiplying the estimated weight α according to the height, the three-dimensional object can be further determined for each cell.

(Other embodiments)
In addition, this invention is not limited only to embodiment mentioned above, In the range which does not deviate from the summary, it is applicable to various embodiment, For example, it can deform | transform or expand as follows.

  In each of the embodiments described above, the bird's-eye view grid G has been set to have a range directed to the front of the host vehicle A. However, a range directed to the side or rear of the host vehicle A can also be set. In this case, a plurality of three-dimensional distance detection units can be provided corresponding to the detection direction.

  Although the example in which the position of the object detection device 1 is made to correspond to the position of the stationary grid S when converting the detected solid object probability corresponding to each cell of the bird's-eye grid G to the stationary grid S has been shown, Assuming that the bird's-eye view grid G is an orthogonal grid based on orthogonal coordinates in which the position of the object detection device 1 is fixed, the bird's-eye view grid G may be processed so as to be superposed corresponding to the position of the stationary grid S after being converted to this orthogonal grid. .

  In the drawings, 1 is an object detection device, 2 is a control device, 2a is a three-dimensional object detection unit, 2b is a road surface detection unit, 2c is a three-dimensional object estimation unit, 2d is a three-dimensional object determination unit, 3 is a three-dimensional distance detection unit, 4 Is a storage unit, and 5 is an output unit.

Claims (4)

A three-dimensional distance detector (3) that acquires at least three-dimensional distance information ahead of the host vehicle;
A three-dimensional object detection unit (2a) that detects the presence or absence of a three-dimensional object based on the three-dimensional distance information for each cell partitioned by a bird's eye grid based on the viewpoint of the three-dimensional distance detection unit;
A road surface detection unit (2b) that detects a pointed cell having distance information between the host vehicle and the solid object as a road surface when the solid object exists;
When there is a pointless area including a pointless cell with no distance information between the host vehicle and the three-dimensional object, an estimated three-dimensional object having the maximum height T1 that can exist in the pointless area is represented by the following formula ( A three-dimensional object estimation unit (2c) estimated by A);
The solid object estimation unit converts at least the estimated solid object cell into a stationary grid cell based on a road surface, and uses the new three-dimensional distance information obtained by the three-dimensional distance detection unit as the vehicle moves. An object detection apparatus comprising: a three-dimensional object determination unit (2d) that detects a solid object when the estimated three-dimensional object cell is repeatedly estimated at the same road surface position when converted to the stationary grid.
T1 = T0 · (re−r1) / re (A)
Where T0 is the height of the three-dimensional distance detection unit, re is the far point distance of the astigmatic area, and r1 is the near point distance of the astigmatic area. The object detection apparatus according to claim 1,
The solid object estimation unit (2c) calculates an estimated solid object having a maximum height Tk that can exist for each of the dotless cells when the dotless region is a plurality of continuous dotless cells by the following formula (B The object detection device estimated by
Tk = T0 · (re−rk) / re (B)
Where T0 is the height of the three-dimensional distance detection unit, re is the far point distance of the astigmatic area, and rk is the near point distance of the target astigmatic cell. In the object detection device according to claim 1 or 2,
The three-dimensional object determination unit (2d) sets an estimated weight according to the height of the estimated three-dimensional object estimated by the three-dimensional object estimation unit, and calculates the estimated three-dimensional object from the three-dimensional distance information associated with the movement of the host vehicle. An object detection device that detects the stationary solid object when a value obtained by adding estimated weights of the object increases. On the computer,
3D distance detection processing for acquiring at least 3D distance information ahead of the host vehicle;
A three-dimensional object detection process for detecting the presence or absence of a three-dimensional object based on the three-dimensional distance information for each cell divided by a bird's eye grid based on the viewpoint of the three-dimensional distance detection unit;
A road surface detection process for detecting a pointed cell having distance information between the host vehicle and the solid object as a road surface when the solid object exists;
When there is a pointless area including a pointless cell with no distance information between the host vehicle and the three-dimensional object, an estimated three-dimensional object having the maximum height T1 that can exist in the pointless area is represented by the following formula ( A three-dimensional object estimation process estimated by A),
The solid object estimation unit converts at least the estimated solid object cell into a stationary grid cell based on a road surface, and uses the new three-dimensional distance information obtained by the three-dimensional distance detection unit as the vehicle moves. An object detection program for executing a three-dimensional object determination process for detecting as a three-dimensional object when the estimated cell of the estimated three-dimensional object is repeatedly estimated at the same road surface position when converted to the stationary grid.
T1 = T0 · (re−r1) / re (A)
Where T0 is the height of the three-dimensional distance detection unit, re is the far point distance of the astigmatic area, and r1 is the near point distance of the astigmatic area.

Images