JP4102885B2 - Parked vehicle detection method and parked vehicle detection system - Google Patents

Description

  The present invention relates to a parked vehicle detection method and a parked vehicle detection system for automatically detecting a parked vehicle.

  Vehicles parked on the road (hereinafter referred to as “parked vehicles”) may cause traffic jams and traffic accidents, and may be a factor that hinders fire fighting activities and emergency activities. For this reason, conventionally, the number of parked vehicles has been investigated by the national and local governments to analyze the impact on traffic congestion and the like.

  Such a survey of parked vehicles is generally performed manually by an investigator on a vehicle or bicycle. However, this investigation method has a problem in terms of investigation accuracy because it takes a lot of work and is inefficient and easily causes mistakes in counting and counting of parked vehicles.

  In order to solve this problem, it is conceivable that the parked vehicle is imaged with a CCD camera or the like, and the captured data is image-processed to identify and count the parked vehicle.

  However, the above-described method for measuring a parked vehicle has a problem in that the detection accuracy of the parked vehicle is not good, although it takes time and cost for image processing. For example, in image processing of captured data, it is difficult to accurately identify a parked vehicle and a building or roadside tree in the background only by simple processing such as edge extraction or binarization processing. In particular, since there is a tendency that a plurality of parked vehicles are continuously parked on the road side, the number of parked vehicles cannot be counted unless individual parked vehicles can be detected.

  Furthermore, image processing becomes even more difficult when acquiring image data of a parked vehicle while moving the survey vehicle in order to increase the speed and efficiency of the survey.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a parked vehicle detection method and a parked vehicle detection system that can efficiently and accurately investigate a parked vehicle.

According to the first main aspect of the present invention, a plurality of line images obtained by performing a line scan of an object on a road a plurality of times in a horizontal direction with a line scan camera mounted on a measurement vehicle traveling on the road are sometimes obtained. A step of obtaining an epipolar plane image (EPI image) having temporal and spatial components by combining along a sequence, and one or more connecting the positions of one or more feature points of the object in the line image in time series a step of the linear extracted as one or more features trajectory of the object in the EPI image, on coordinates of the component axes of time and space in the EPI image, the step of calculating an inclination of the extracted feature trajectory If, based on the slope of the calculated characteristic trace and the moving velocity of the measuring vehicle, a step of calculating according to a predetermined formula the depth distance to feature points of the object, feature points of the computed object Parked vehicle detection method characterized by comprising the step of determining the presence or absence of a parked vehicle by a depth distance is compared with a predetermined threshold in is provided.

According to such a configuration, since the EPI image (imaging data) acquired by a plurality of line scans is analyzed to discriminate the parked vehicle, the parked vehicle can be reliably discriminated. Further, by acquiring data around the parked vehicle at the same time, it is possible to provide data useful for road administration such as the regulation of the parked vehicle.

Moreover, according to the second main aspect of the present invention, a line scan camera mounted on a measurement vehicle traveling on the road acquires a plurality of line images obtained by performing a line scan of an object on the road a plurality of times in the horizontal direction. An image data acquisition unit, a road imaging unit that acquires a horizontal epipolar plane image ( EPI image ) by synthesizing a plurality of line images of the object in time series, and an object in the line image a feature trajectory extraction unit that extracts one or more straight line connecting one or more of the positions of feature points in chronological as one or more features trajectory of the object in the EPI image, the component of time and space in the EPI image Based on the inclination calculation unit that calculates the inclination of the extracted feature trajectory on the coordinate axis, and the calculated feature trajectory and the moving speed of the measurement vehicle, the feature point of the target object is calculated. A distance calculation unit that calculates the depth distance in accordance with a predetermined expression, and a parked vehicle presence / absence determination unit that determines the presence / absence of a parked vehicle by comparing the calculated depth distance to the feature point of the target object with a predetermined threshold. A parked vehicle detection system is provided.

  According to such a structure, the parked vehicle detection system which can implement | achieve the parked vehicle detection method in the above-mentioned 1st main viewpoint suitably can be obtained.

As described above, according to the present invention, it is possible to obtain a parked vehicle detection method and a parked vehicle detection system that can efficiently and accurately investigate a parked vehicle.

  In particular, it is possible to obtain a parked vehicle detection method and the like that can determine and count a parked vehicle while acquiring imaging data from a traveling survey vehicle, and that can efficiently survey the parked vehicle.

  The other features and remarkable effects of the present invention can be understood more clearly by referring to the description of the following embodiments of the present invention and the attached drawings.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the following, first, an embodiment serving as a reference of the present invention will be described, and then an embodiment of the present invention will be described.
(Reference embodiment)
FIG. 1 is a schematic explanatory diagram of a parked vehicle detection system according to this embodiment. The parked vehicle detection apparatus denoted by reference numeral 1 in this figure acquires imaging data including a parked vehicle 3 on the road by the laser scanner 2 and is mounted on a measurement vehicle 5 traveling on the road 4. The parked vehicle detection device 1 detects the side surface of the parked vehicle 3 from road data captured by the laser scanner 2 based on distance information to the parked vehicle 3 and an object such as a background roadside tree 6 or a building 7, that is, depth. Data is extracted. In this embodiment, the acquisition of the imaging data, the determination of the parked vehicle 3, the counting, and the output are executed in real time on the measurement vehicle 5 that travels. As a result, the parked vehicle can be investigated efficiently.

  Here, a schematic configuration of the laser scanner 2 which is a road imaging unit in the present embodiment will be described with reference to FIG.

  The laser scanner 2 mainly includes a laser light transmitting unit 8 that transmits laser light to an object, and a laser light receiving unit 9 that receives laser light reflected from the object. The laser beam sending unit 8 includes a semiconductor laser diode 10, a pair of prisms 11 and 11, a sending lens 12, a rotatable polygon mirror 13, and a scan line adjusting mirror 14. The laser beam receiver 9 includes a plurality of receiving mirrors 15, 15, a receiving lens 16, and a photodiode 17. Reference numeral 18 denotes a window for transmitting and receiving laser light.

  In such a configuration, when laser light is oscillated from the semiconductor laser diode 10, a one-dimensional scan line is drawn in the horizontal direction by the rotation of the polygon mirror 13, and line scanning becomes possible. Further, by adding the rotation of the mirror 14, the scan line is changed in the vertical direction by a minute angle, and two-dimensional area scanning becomes possible. When performing a line scan in the vertical direction, the laser scanner 2 may be installed with an inclination of 90 degrees.

  As a method of measuring the distance to the object using the laser scanner 2, there are, for example, a time difference measurement method and a phase difference measurement method. Here, an example of the time difference measurement method will be described with reference to FIG. This method is a method for obtaining the distance to an object by directly measuring the time-of-flight (TOF) of the pulse laser beam. In this measurement method, the distance D [m] to the object is obtained from the measured time difference Δt [sec] by the following equation.

2D = cΔt (where light velocity c = 3.0 × 10 ² [m / sec])
When the time difference measurement method is employed, the speed of light is fast, and thus a short time interval must be accurately measured in order to improve measurement accuracy. For example, a time resolution of 6.7 × 10 ⁻⁹ sec (= 6.7 ns) is required when the measurement accuracy is 1 m, and a time resolution of 0.067 ns is required at 1 cm. Note that the measurement accuracy of this method has a characteristic that it does not depend on the size of the measurement distance in principle.

  Combining such distance measurement with the two-dimensional area scan mechanism described with reference to FIG. 2 enables three-dimensional measurement of an object on the road. At this time, the three-dimensional shape of the object is represented as a set of points having highly accurate three-dimensional position information, that is, range points.

  An example in which the parked vehicle 3 is represented by this range point is shown in FIG. Innumerable points in the figure are range points. As is apparent from this figure, the vehicle body of the parked vehicle 3 is shown as a portion (set) where the range points are dense. By extracting such a set, the vehicle body as well as the presence of the parked vehicle 3 is shown. Can be determined.

  In addition, behind the parked vehicle 3 (y-axis direction), there is a region where no range point appears, that is, a region (shadow) where the laser beam is shielded by the parked vehicle 3. The presence or absence of the parked vehicle 3 can also be easily determined by extracting this shielding area.

Next, the configuration of the parked vehicle detection device 1 will be described with reference to the block diagram of FIG.
The device 1 is provided in a computer system mounted on the measurement vehicle 5. A data storage unit 25 and a program are connected to a bus 24 to which a CPU 20, a RAM 21, a communication device 22 such as a modem and an input / output device 23 are connected. And a storage unit 26.

  The data storage unit 25 stores an imaging data storage unit 27 that stores data captured by the laser scanner 2, and parked vehicle data storage that stores side data of the parked vehicle 3 determined by the parked vehicle presence / absence determination unit 36 described later. And a measurement condition storage unit 29 for storing the measurement conditions / measurement environment and the like of the parked vehicle 3. Here, the measurement condition storage unit 29 stores, for example, the date and time of measurement, the weather condition at the time of measurement, the average speed of the measurement vehicle 5, the set threshold value, and the like.

  The program storage unit 26 receives the image data on the road imaged by the laser scanner 2 in addition to the main program 32, and stores the image data acquisition unit 33 in the imaging data storage unit 27. Measured from the data extraction unit 34, the shielding region extraction unit 35, the parked vehicle presence / absence determination unit 36, the number of parked vehicle units calculation unit 37, the parking region occupancy rate calculation unit 38, and the number of rotations of tires of the measurement vehicle 5, for example The speed / distance information acquisition unit 39 that acquires speed and distance conditions and stores them in the measurement condition storage unit 29, and the input / output of processing results such as the calculated number of parked vehicles 3 and parking area occupancy And a processing result output unit 40 for outputting to a display or the like of the device 23.

  The side data extraction unit 34 extracts side data of the parked vehicle 3 parked on the road from image data obtained by imaging an object on the road, and has the configuration shown in FIG.

  That is, the side data extraction unit 34 includes an imaging data pre-processing unit 41 that performs pre-processing such as noise removal of the acquired imaging data, and information on the height of the range point included in the imaging data (z axis in FIG. 4). A height calculating unit 42 for calculating the height of the object based on the value of the object), and a height for setting a threshold of the height of the object based on at least one of a general vehicle height and a tire diameter of the vehicle A height threshold value setting unit 43, and a height data extracting unit 44 that randomly extracts height information for each predetermined number of parked vehicles 3 or for each predetermined scanning period of laser light determined by the parked vehicle presence / absence determining unit 36. A threshold updating unit 45 that dynamically updates the set threshold based on a plurality of extracted height information, and a predetermined number of frames in which the calculated height of the target object is equal to or greater than the predetermined threshold. Its consecutive frames A side data output unit 46 that outputs the side data of the parked vehicle 3, and a reference frame number setting unit 47 that sets a predetermined number of reference values in which the frames continue based on the moving speed of the moving body. .

  Here, the imaging data pre-processing unit 41 mainly converts the sensor coordinate system of the imaging data acquired from the laser scanner 2, restores the three-dimensional shape of the imaged object, and the height calculation unit 42. Each process of smoothing the calculated vehicle body side surface point set is performed.

  Further, the height data extraction unit 44 extracts the height data at random as described above and sends the height data to the threshold update unit 45. As a result, the threshold value can be dynamically updated to improve detection accuracy, and the extraction of side data can be verified by performing the side data extraction process again based on the updated threshold value. .

  The details of each process performed by the imaging data preprocessing unit 41 and the height calculation unit 42 will be described later.

  Next, the shielding area extracting unit 35 shown in FIG. 5 extracts an area (occlusion) where the laser beam is shielded by the object from the imaging data (see FIG. 4). In the present embodiment, the shielding area extraction unit 35 has a function of complementing the side data extraction unit 34. That is, when the object is black, the reflectance of the laser light is extremely low, and therefore there is a possibility that side data cannot be extracted for the black parked vehicle 3. Therefore, the discrimination accuracy is improved by using the discrimination of the parked vehicle 3 by extracting the shielding area together. The function and the like of the shielding area extraction unit 35 will be described in detail later.

  The parked vehicle presence / absence discriminating unit 36 discriminates the presence / absence of the parked vehicle 3 based on the side data of the parked vehicle 3 extracted by the side surface data extracting unit 34 or the shielding area data extracted by the shielding area extracting unit 35. is there.

  The parked vehicle number calculation unit 37 counts the number of parked vehicles 3 determined by the parked vehicle presence / absence determination unit 36 in real time.

  The parking area occupancy rate calculation unit 38 divides the number of frames determined by the parked vehicle presence / absence determination unit 36 that the parked vehicle 3 is present by the total number of frames of the imaging data, so that the parking area can be parked. The ratio of parked vehicles is calculated. A specific concept of this occupation ratio is shown in FIG.

  In this figure, the number of frames occupied by the parked vehicle 3 is indicated by the right-down diagonal line, and the total number of frames to be measured is indicated by the right-up diagonal line. As is apparent from this figure, the fixed value (for example, 3 m) of the vehicle length is not uniformly calculated by multiplying the number of parked vehicles 3, but the number of frames of the parked vehicle 3 (that is, the total length of the vehicle) ) Can be calculated more accurately.

  In this way, by providing a new index called the occupation ratio in the parking area, it can be expected to be effectively used for solving various problems in road administration such as elimination of traffic congestion.

  Each of the above components is actually a computer software program installed in a computer system or a subroutine in one program. Each function of the present invention is exhibited by being called and executed on the RAM 21 by the CPU 20.

  Next, each process performed by the imaging data pre-processing unit 41 will be described with reference to FIGS.

First, the transformation of the sensor coordinate system will be described.
With the data captured by the laser scanner 2, the following information can be obtained for one range point.

Frame number: m (number of the frame to which the point belongs; m = 1, 2, 3, L)
Scan number: n (number of the point in the frame; n = 1, 2, 3,... 165)
Distance to point: r
Swing angle: α (constant value; α = 0 °)
Scan angle: β (−50 ° ≦ β ≦ + 20 °)
Measurement time: t (only when n = 1)
Here, in the present embodiment, a line scan for scanning the object in the vertical (longitudinal) direction is performed, and the swing angle α indicating the measurement angle in the horizontal direction is fixed at 0 °. Further, the scan angle β represents a measurement angle in the vertical direction. As shown in FIG. 8, these measurement angles are set with reference to the distance between the measurement vehicle 5 and the parked vehicle 3 to be measured, the vehicle height, the installation height of the laser scanner 2, and the like.

  In the present embodiment, the object on the road is scanned by the laser scanner 2 while the measuring vehicle 5 travels, and the position of the laser scanner 2 changes from moment to moment. Therefore, it is necessary to accurately represent the position of the range point in the real space by converting “the distance r to the point, the swing angle α and the scan angle β”, which are values on the “sensor coordinate system”, into the world coordinate system, respectively. There is. Such transformation of the coordinate system is performed in the process shown in FIG. FIG. 10 shows the relationship between the world coordinate system (p) and the sensor coordinate system (p ′).

  Since the range points all belong to a specific scan line and have information indicating the number of points scanned among them, m and n ID tags are added. Therefore, the three-dimensional shape of the object can be expressed as follows.

  The state of FIG. 4 described above restores the three-dimensional shape of the parked vehicle 3 by performing the coordinate conversion as described above. In other words, in this process, the data acquired by the laser scanner 2 has distance information from the measurement point to the object, and is described by a polar coordinate system model. By converting this data into an orthogonal coordinate system model, the three-dimensional shape is restored.

  The imaging data preprocessing unit 41 also smoothes the vehicle body side point set. This smoothing process performs filtering processing with a predetermined width on a later-described height curve created by the height calculation unit 42, and applies a predetermined weighting coefficient to reduce noise included in the height curve. To be removed.

  Next, the step of calculating the height of the object included in the imaging data by the height calculation unit 42 will be described in detail.

  First, the height calculation unit 42 extracts a set (B) of points indicating the side surface of the vehicle body from the imaging data according to the following expression.

  In this equation, among the points that do not belong to the road surface point set (A) extracted by the following equation, a range point whose y coordinate is within an appropriate range is regarded as belonging to the vehicle body side surface point set (B). Note that the following equation is also used when extracting the shielding area.

An example of the “appropriate range” is shown in FIG. In this figure, in order to avoid complication, a portion where the range points are dense is indicated by a diagonal line rising to the right, and a portion where the density of the range points is small is indicated by a diagonal line rising to the left. In this example, the minimum value is set to a fixed value of 1.0 m, and the maximum value is set to a depth position of y _peak and the maximum value is set to 1.0 m deeper than that. This y _peak can be defined as “y-coordinate that takes a histogram of the number of points along the y-axis direction in the point set that does not belong to the road surface point set and gives the first maximum value”. As a result, y _peak exists in the vicinity of the vehicle body right side of the parked vehicle parked in parallel, and the point set on the side surface of the vehicle body can be accurately extracted. In addition, it is preferable to extract a region including the side surface of the parked vehicle in consideration of the measurement accuracy (0.5 m or the like) of the laser scanner 2. In this figure, the dense part of the range points appearing on the back side of the parked vehicle is considered to be the background of the guardrail or the wall of the building.

  The height curve shown in FIG. 12 is obtained by connecting the points having the maximum coordinates for each scan line in the vehicle body side point set (B) extracted in this way. For this height curve, the imaging data pre-processing unit 41 removes high-frequency noise with a smoothing filter to obtain a line representing a ridge line (contour line) on the side surface of the vehicle body. As an example, the filter width is set to 3 (near the front and back 1), and each weighting coefficient is set to 1.

  The main variable of the height curve is a scan line number n, which can be expressed by the following equation.

  FIG. 12A shows a height curve before smoothing processing by the imaging data preprocessing unit 41, and FIG. 12B shows a height curve after smoothing processing. It can be seen from the imaging data shown in this figure that three pieces of vehicle side surface data are extracted.

  Next, the process in which the parked vehicle presence / absence determining unit 36 determines the presence / absence of a parked vehicle based on the height curve will be described in detail.

  This parked vehicle presence / absence discriminating unit 36 gives a predetermined threshold to the height curve shown in FIG. 12, and determines from the intersection of the height curve to the intersection as a portion where there is a candidate for the parked vehicle or a portion where it does not exist. Is.

If the threshold given to the height curve is _zth , the vehicle candidate presence determination is expressed by the following equation.

Here, E _m is a 0/1 variable indicating whether a candidate for a parked vehicle exists in the m-th frame. 1 means that there is a vehicle candidate and 0 means that it does not exist.

Further, the threshold value z _th can be given by the following equation, for example.

  The fixed value of 0.5 m is set by the height threshold setting unit 43 in consideration of the vehicle main body located at a position higher than the tire diameter of the parked vehicle 3.

Wherein E _m is the information concerning only the one scan line. Therefore, a pattern in which 1 continues like “0111... 1110” is searched from the sequence of E ₁ , E ₂ , E ₃ ,. Also, to remove the influence of noise and outliers, 1 or 0 of the width within 4 as previously "011110" and "100001" is kept correct the _{E m} all converted to 0 or 1.

In the search pattern, identified as one of the parked vehicle what width is equal to or greater than F _m of "111L111". The value of F _m is set by the reference frame number setting unit 47 according to the traveling speed v of the measurement vehicle 5. Here, a vehicle with a length of 2 m or more is identified as one vehicle, and given by the following equation.

  By performing such processing, a threshold line can be drawn at an appropriate position as shown in FIG. 12B, and the parked vehicle can be accurately identified based on the height curve.

  Hereinafter, the detailed function of this parked vehicle detection apparatus 1 is demonstrated with reference to FIG. 13 with an actual operation | movement. In addition, S1-S12 of these figures is a code | symbol which shows a process order, and respond | corresponds to step S1-S12 of the following description.

  First, the imaging data acquisition unit 33 acquires imaging data obtained by imaging the road by the laser scanner 2 and stores it in the imaging data storage unit 27 (step S1).

  Next, for the acquired imaging data, the imaging data preprocessing unit 48 executes a predetermined preprocessing. Specifically, as described above, the “sensor coordinate system” is converted into the world coordinate system (step S2), and the three-dimensional shape of the object is restored by the point set with ID tag as shown in FIG. 11 (step S2). S3).

  Next, the height calculator 42 extracts a point set on the side surface of the vehicle body (step S4). Since the extracted point set includes a lot of noise, the imaging data preprocessing unit 48 removes the noise by the smoothing process (step S5).

  Next, the height calculation unit 42 calculates the height curve of the object (step S6), and the height threshold setting unit 43 sets a threshold for the object height curve based on the vehicle height and the tire diameter. A line (0.5 m in this example) is calculated (step S7). In this state, the side data output unit 46 extracts a frame whose height is larger than the threshold line and sends it to the parked vehicle presence / absence determining unit 36 (step S8). The extracted height data is a vehicle candidate.

The parked vehicle presence / absence discriminating unit 36 that has received the extracted height (vehicle candidate) data determines whether or not “E _m = 1” frames in which the height of the object is greater than the threshold line continue for a predetermined number or more. For non-continuous frames, correction is made to “E _m = 0” (step S9). As a result of the correction, a frame in which a predetermined number of “E _m = 1” continues is determined as the parked vehicle 3 (step S10). The parked vehicle number calculating section 37 counts and outputs the number of parked vehicles 3 (step S11). Further, as shown in FIG. 7, the parking area occupancy rate calculation unit 38 calculates the occupancy rate of the parked vehicle 3 based on the number of frames determined to be the parked vehicle 3 (step S <b> 12).

Next, the shielding area extraction unit 35 will be described.
In the present embodiment, as described above, the presence / absence of the parked vehicle 3 is determined based on the side data extracted by the side data extraction unit 34 and the shielding area (a set of road surface points) of the laser light extracted by the shielding region extraction unit 35. It has the feature in the point which discriminate | determines.

  As shown in FIG. 14, the shielding region extraction unit 35 according to this embodiment is based on the imaging data preprocessing unit 50 and the depth point depth information (value y in FIG. 4) included in the imaging data. A depth calculation unit 51 for calculating the depth of the object, a depth threshold setting unit 52 for setting a threshold of the depth of the object based on the full width of the vehicle, a depth data extraction unit 53, and a threshold update unit 54 When a predetermined number of frames whose depth of the target object is equal to or less than a predetermined threshold continues, a shielding area output unit 55 that outputs the consecutive frames as a shielding area by the parked vehicle 3 and a reference frame number setting unit 56 are provided. ing.

  The depth calculation unit 51 first extracts a set (A) of points indicating the road surface from the imaging data by the following equation. This road surface point set extraction is performed by paying attention to the vertical position (height from the road surface) of the range point. In this example, the distance between the tires before and after the parked vehicle (the bottom of the vehicle body) reaches the back side of the vehicle body side, so that the height is based on the range point at a position higher than the tire diameter. Was set to 0.5 m. Also in this case, it is preferable to determine a region for extracting a road surface point set in consideration of the measurement accuracy of the laser scanner 2.

  Next, a depth curve shown in FIG. 15 is created from the extracted road surface point set. Specifically, it is obtained by connecting the points with the maximum y coordinate for each scan line in the extracted set of road surface points. This depth curve is expressed as follows, with scanline number n as the main variable.

  This depth curve is created based on a range point with a small position away from the laser scanner 2 and contains a lot of noise. Therefore, the depth curve is smoothed by filtering by the imaging data preprocessing unit 50. Specifically, the road surface data distribution region means a visible region on the road surface by the laser, and the boundary in the depth (y-axis) direction can be configured from the maximum value of y for each scan. By removing high-frequency noise from the boundary line in the depth direction using a smoothing filter, a depth curve representing a shielded portion by a road parked vehicle can be suitably acquired. As an example, the width of the filter is set to 9 (near 4 front and rear), and each weighting factor is set to 1.

  Thereby, the depth curve of FIG. 15 is obtained. FIG. 15A shows a depth curve before smoothing processing by the imaging data preprocessing unit 50, and FIG. 15B shows a depth curve after smoothing processing. From the imaging data shown in this figure, it can be seen that three pieces of shielding area data are extracted.

  Further, when the shielding area data in this state is sent to the parked vehicle presence / absence determining unit 36, the parked vehicle presence / absence determining unit 36 gives a predetermined threshold to the depth curve shown in FIG. To the intersection is determined to be a portion where a candidate for a parked vehicle exists or a portion where no candidate exists.

When the threshold value that gives the depth curve and y _th, presence determination of vehicle candidates are expressed by the following equation. This threshold value is set by the depth threshold value setting unit 52.

Here, E _m is a 0/1 variable indicating whether a candidate for a parked vehicle exists in the m-th frame. 1 means that there is a vehicle candidate and 0 means that it does not exist.

Further, the threshold value y _th can be given by the following equation, for example.

  Thus, the discrimination accuracy of a parked vehicle can be improved by setting a threshold value based on data that is identified or highly likely to be identified as a parked vehicle. Moreover, the fixed value of 1.5 m takes into consideration that the full width of the parked vehicle 3 is normally about 3 m.

By applying the threshold line set in this way (about 5.5 m in this example) and correcting _Em as in the first embodiment, a parked vehicle as shown in FIG. Three shielding areas can be output.

  Hereinafter, detailed functions of the system 1 according to this embodiment will be described together with actual operations with reference to FIG. Note that S13 to S26 in these figures are codes indicating the processing order and correspond to steps S13 to S26 in the following description. Further, detailed description of the processing steps common to the first embodiment described above is omitted.

First, the imaging data acquisition unit 33 acquires imaging data and stores it in the imaging data storage unit 27 (step S13). Next, for the acquired imaging data, the imaging data preprocessing unit 48 executes a predetermined preprocessing (steps S14 and S1).
5).

  Next, the depth calculation unit 51 extracts a set of road surface points (A area) based on the height information of the range points of the imaging data (step S16). Further, the imaging data preprocessing unit 50 smoothes road surface point information and removes noise (step S17).

  Next, the depth calculation unit 51 calculates the depth curve of the object (step S17), and the depth threshold setting unit 52 sets the threshold of the depth curve based on the full width of the parked vehicle (steps S18 and S19). . In this state, the shielding area output unit 55 extracts a frame whose depth is smaller than the threshold line and sends it to the parked vehicle presence / absence determination unit 36 (step S21). The extracted height data is a vehicle candidate.

  The parked vehicle presence / absence discriminating unit 36 determines whether or not a predetermined number or more of frames whose depth is lower than the threshold line is continuous, and corrects the frames that are not continuous (step S22).

  Next, the parked vehicle presence / absence discriminating unit 36 discriminates the parked vehicle 3 based on the extracted road surface point data (step S23). Based on the determined result, the number of parked vehicle number calculation unit 37 outputs the number of parked vehicles, and the parking area occupancy rate calculation unit 38 calculates the occupancy rate of the parked vehicle 3 (steps S24 and S25).

(Embodiment of the present invention)
Next, an embodiment of the present invention will be described with reference to FIGS. This embodiment is characterized in that a three-dimensional epipolar plane image (EPI image) is acquired using a line scan camera, and a parked vehicle is determined based on the inclination of the characteristic locus of the acquired image data.

  As shown in FIG. 17, the parked vehicle detection device 60 according to this embodiment acquires an epipolar plane image (EPI image) by performing line scanning of an object on the road a plurality of times in the horizontal direction from the measurement vehicle 5 traveling on the road. A line scan camera 61, a road imaging unit 62 that acquires an EPI image in a horizontal direction by synthesizing EPI images of the object in time series, and at least one of the objects related to the object from the acquired EPI images Based on the characteristic trajectory extraction unit 63 that extracts the characteristic trajectory, the inclination calculation unit 64 that calculates the inclination of the characteristic trajectory of the extracted object, and the measured characteristic trajectory and the traveling speed of the measurement vehicle 5 A distance calculation unit 65 that calculates the distance from the vehicle 5 to the object, and a parked vehicle presence / absence determination unit 66 that determines the presence or absence of the parked vehicle 3 based on the calculated distance to the object. Eteiru.

  Image data of the actual parked vehicle 3 is processed in FIG. 18A, and an image acquired by the line scan camera 61 is processed in FIG. 18B by the feature locus extraction unit 63, the inclination calculation unit 64, and the like. The results are shown in FIG.

  First, the image of FIG. 18B is generated by arranging a line image in the right direction horizontally each time the scanning is repeated. From this, it can be considered that the direction from left to right is time, and the vertical direction is a spatio-temporal image that means space, that is, an epipolar plane image (EPI image).

  In this image, the trajectory 67 of the feature point of the object appears as a straight line descending to the right. By processing this image, a feature path can be extracted and the depth distance from the inclination of the straight line to the feature point can be obtained.

  Hereinafter, specific image processing steps will be described.

(Pretreatment process)
First, the spatiotemporal image is divided by a certain number of lines in the time direction without overlapping. By detecting an edge of each divided image using the Canny edge detection algorithm, the image is binarized as shown in FIG. From this figure, it can be seen that the straight line appears clearly.

(Linear extraction process)
Next, in order to extract a straight line in the obtained binary image, a Hough transform process is performed. The straight line in the divided image can be uniquely expressed by r = OR and the normal vector angle φ, where R is a perpendicular line drawn from the origin O to the straight line as shown in FIG. FIG. 21 shows a Hough space image obtained by applying the Hough transform to the upper divided image. There is a peak near r = −500 and φ = 145 degrees, which means the strongest feature trajectory. Therefore, the inclination of the feature locus in the divided image is 145−90 = 55 °.

(Depth calculation process)
For the extracted slope m of the straight line, the depth distance L to the corresponding feature point is calculated according to the following equation.

  Therefore, the depth distance L and the slope m of the straight line in the EPI image are proportional. Therefore, the Hough transform is applied to an image divided by 200 lines in the time direction, and only the straight line corresponding to the maximum peak in the Hough space image obtained for each divided image is extracted. This straight line means the feature locus with the strongest edge in the EPI. FIG. 18C is a graph in which the angle φ in the normal direction of the extracted straight line is plotted according to the time series of the divided images. From this figure, it can be seen that the normal direction angle φ = 145 °, that is, the depth distance of the feature trajectory having the inclination 145-90 = 55 ° in the EPI is the shortest. This corresponds to the depth distance of the parked vehicle.

(Threshold setting process)
The time change of the normal direction angle of the straight line extracted from the EPI image is obtained as a graph as shown in FIG. In addition, since the normal direction angle and the depth distance are proportional, it is easy to use the threshold value as shown in the graph of FIG. It can be seen that the threshold φ _{th at} this time depends on the speed V of the camera (= measurement vehicle) and the depth distance L _th as shown in the following equation.

Although the speed V of the measurement vehicle 5 can be measured, the distance L _th between the measurement vehicle 5 and the parked vehicle 3 depends on the measurement environment, particularly the width of the road lane and the lateral movement locus of the measurement vehicle 5. It is difficult to specify its value. Accordingly, it is shown lane width, full width sedan type of vehicle as an example of the parked vehicle, and a step of calculating the L _th based on the entire width of the measuring vehicle 5 in FIG. 22.

(Normalization)
In this example, since the size of the divided image is an image having an aspect ratio of about 1: 7, the length of the vertical straight line is considerably shorter than the horizontal even for a straight line having a similarly strong edge. Become. Therefore, the index of the strength of the straight line in the extraction is normalized as a relative strength with respect to the length of the straight line. Thereby, noise can be reduced.

(Effect of embodiment)
According to each embodiment described above, the following effects can be obtained.

  1) A side data extraction step for extracting side data of a parked vehicle from data obtained by imaging an object on the road, a parked vehicle presence / absence determination step for determining the presence / absence of a parked vehicle based on the extracted side data, A road imaging process is performed in which a laser beam is transmitted from a traveling vehicle to the object while scanning and the reflected laser beam is detected to acquire imaging data of the object. Thereby, the presence or absence of the parked vehicle can be determined based on the side data of the parked vehicle extracted from the imaging data. Further, by referring to the side data, there is little influence due to the alignment state of the parked vehicles, and the detection accuracy can be improved. Furthermore, by scanning with laser light, it is possible to acquire imaging data without being affected by weather, shadows, outdoor light, and the like. In this method, since the distance to the object can be directly measured, the cost and time required for analyzing the acquired data can be reduced. Furthermore, since the side of the vehicle body having a high density of range points reflecting the laser beam is referred to, the influence of noise can be minimized, and the detection accuracy can be further improved.

  2) Counting the number of parked vehicles while acquiring imaging data by laser light by executing the parked vehicle number calculating step of counting and outputting the number of parked vehicles determined in the parked vehicle presence / absence determining step Will be able to.

  3) In the road imaging step, the laser beam is scanned in the horizontal direction, and the scan line belonging to each frame of the imaging data is overlapped with the scan line belonging to one or more consecutive frames. Epipolar plane distance image data having information on the distance to the object is obtained by scanning or further scanning the object on the road with the laser beam a plurality of times in the horizontal direction. Thereby, imaging data useful for discrimination of a parked vehicle can be obtained by line scanning.

  4) Imaging data having distance data to an object on the road based on the time difference from the transmission of the laser light to the detection of the reflected light or the phase difference between the irradiated laser light and the detected reflected light in the road imaging step. By acquiring the vehicle side data, the vehicle side data can be reliably extracted based on the acquired imaging data.

  5) In the side surface data extraction step, a height calculation step of calculating the height of the target object based on information on the height of the range point included in the imaging data, and the calculated height of the target object is predetermined. A side data output step of outputting a frame equal to or greater than the threshold value as side data of the parked vehicle is executed. Further, in the side surface data extraction step, a height threshold value setting step for setting a threshold value of the height of the object based on the vehicle height of the vehicle is executed. Further, in the height threshold value setting step, a height information extracting step for extracting height information for each predetermined number of parked vehicles or for each predetermined scanning period of the laser beam determined in the parked vehicle presence / absence determining step, and extraction And a threshold updating step for updating the set threshold based on the plurality of height information. Further, in the side data output step, when a predetermined number of frames having an object height equal to or higher than a predetermined threshold value continue, the continuous frames are output as side data of the parked vehicle. With such a configuration, it is possible to set / update a threshold value when extracting side surface data from imaging data to an appropriate value.

  6) A depth calculation step of calculating the depth of the object based on the depth information of the range point included in the imaging data in the side surface data extraction step, and the calculated depth of the object is equal to or less than a predetermined threshold value. The side data output process for outputting the frame as side data of the parked vehicle is executed. Also, in this side surface data output step, a depth threshold setting step for setting a depth threshold of an object based on the full width of the vehicle is executed, or a predetermined number of frames having a depth of the object equal to or less than a predetermined threshold are consecutive In addition, the continuous frame is output as the side data of the parked vehicle, or the reference frame number setting step of setting a predetermined number of reference values in which the frame continues is executed based on the moving speed of the vehicle. I made it. Further, in the depth threshold setting step, a depth information extracting step for extracting depth information for each predetermined number of parked vehicles or for each predetermined scanning period of the laser light determined in the parked vehicle presence / absence determining step, and a plurality of extracted And a threshold value updating step for updating the set threshold value based on the depth information. With such a configuration, it is possible to extract side surface data from this imaging data based on depth information included in the imaging data.

  7) further executing a shielding region extraction step of extracting a region (occlusion) where the laser beam is shielded by the object from the imaging data, and in the parking vehicle presence / absence determination step, the extracted side data of the parked vehicle or By determining the presence / absence of a parked vehicle based on at least one of the shielding area data, the parked vehicle can be identified based on the shielding area even when it is difficult to extract side data.

  8) A parking area occupancy ratio in which the number of frames determined to have a parked vehicle in the parked vehicle presence / absence determining step is divided by the total number of frames of the imaging data to calculate the ratio of the parked vehicle in the parking area By executing the calculation step, it is possible to provide objective information that is useful for the regulation of parked vehicles.

  9) A parking vehicle detection device mounted on a measurement vehicle that travels on the road, a road imaging unit that irradiates the object with laser light, receives laser light reflected from the object, and acquires imaging data; A side data extraction unit that extracts side data of a parked vehicle from the acquired imaging data, a parked vehicle presence / absence judgment unit that discriminates the presence or absence of a parked vehicle based on the extracted side data, and counts the number of parked vehicles And a parked vehicle count unit for outputting. Thereby, each process of acquisition of imaging data, discrimination | determination of a parked vehicle, and count of the discriminate | determined parked vehicle can be performed in real time, drive | working with a measurement vehicle.

  In addition, this invention is not limited to each above-mentioned embodiment, A various deformation | transformation is possible in the range which does not change the summary of invention.

  For example, in the above-described embodiment, the parked vehicle detection device 1 mounted on the measurement vehicle 5 is mounted and the parked vehicle 3 is detected in real time while traveling on the road. However, the present invention is not limited to this. For example, the measurement vehicle 5 may be equipped with only the laser scanner 2 and the imaging data storage unit 27, and the acquired imaging data may be processed collectively at a later time. As a result, the equipment mounted on the measurement vehicle 5 can be simplified, and it is not necessary to increase or decrease the traveling speed of the measurement vehicle 5 in consideration of the processing speed of the imaging data.

  Moreover, the imaging data acquired by the laser scanner 2 may be transmitted to the server of the information processing center as needed to process the imaging data acquired by the plurality of measurement vehicles 5 in a concentrated manner.

  Moreover, this invention is applicable not only to the detection of the parked vehicle 3 on a road but to the detection of the parked vehicle 3 in various parking lots. In addition, the moving body on which the laser scanner or the like is mounted is not limited to the measuring vehicle 5, and may be a measuring device that can move on a track previously laid in a predetermined area such as a parking lot or a roadside of a specific road. In this case, it is preferable to periodically detect parked vehicles in the area by moving the measuring device at a predetermined cycle.

  Furthermore, the type of sensor that acquires the imaging data, the analysis method of the imaging data, the algorithm, and the like are not limited to the above-described examples, and can be variously changed. The scanning direction of the laser scanner may be any of a vertical direction, a horizontal direction, and an oblique direction.

  Even when the measuring vehicle 5 travels at a non-constant speed, the self-position of the measuring vehicle 5 (laser scanner 2 or the like) is simultaneously recorded using GPS (global positioning system) or INS (inertial navigational system). By doing so, the detection accuracy can be maintained.

  It is also possible to acquire an EPI image as in the embodiment of the present invention by using the laser scanner 2 of the reference embodiment described above. For example, as shown in FIG. 23, horizontal scan lines are arranged vertically like layers along the time axis like a line scan camera (FIG. 23A). FIG. 23 (b) shows that this layer is slid horizontally according to the traveling speed. When this is projected onto a plane, the images shown in FIGS. From this figure, it can be seen that there are four parked vehicles. In this method, by applying the EPI analysis method to the re-processing to derive the image of (b), even if the moving speed of the laser scanner (measurement vehicle) is unknown, appropriate alignment is performed. It is something that can be done.

  Below, the detection result of the parked vehicle performed using the parked vehicle detection apparatus demonstrated as the said embodiment is demonstrated.

(Example 1)
First, an example in which the presence or absence of a parked vehicle is determined by extracting vehicle side data or a shielding area from data captured by a laser scanner mounted on a measurement vehicle will be described.

FIG. 24 shows the measurement conditions, and FIG. 25 shows the measurement results.
As is apparent from these drawings, the presence of the parked vehicle can be accurately determined from the data captured by the laser scanner mounted on the traveling measurement vehicle. In addition, considering that it takes only a few seconds to analyze the image data captured by the laser scanner, it is easy to understand that it is possible to discriminate parked vehicles on the roadside and count in real time while traveling at high speed. it can. Furthermore, good measurement results were obtained both in the method of extracting the vehicle side surface data and the method of extracting the shielding area.

(Example 2)
Next, an example will be described in which a feature trajectory of a vehicle is extracted from an EPI image captured by a line scan camera mounted on a measurement vehicle, and the presence / absence of a parked vehicle is determined based on its inclination.

FIG. 26 shows the measurement conditions, and FIG. 27 shows the measurement results.
As can be understood from these figures, almost all of the parked vehicles were detected, but there were some that were erroneously detected. There was no difference in accuracy depending on the running speed.

  Causes of erroneous detection include, for example, reflected light from a parked vehicle, reflection in the surroundings of the parked vehicle, and a parked vehicle having a special structure. These causes can be easily removed by attaching a hood to the lens, using a circular deflection filter, or using a plurality of line scan cameras.

BRIEF DESCRIPTION OF THE DRAWINGS Schematic explanatory drawing of the parked vehicle detection system concerning embodiment used as the reference of this invention. The schematic block diagram of a laser scanner. The figure for demonstrating the time difference measuring method. The figure showing the parked vehicle by the range point. The block diagram which shows schematic structure of a parked vehicle detection apparatus. The figure which shows schematic structure of a side data extraction part. The figure for demonstrating the idea of a parking area occupation rate. The figure which shows scan angle (beta) (measurement angle of a perpendicular direction). The figure which shows the conversion process of a coordinate system. The figure which shows the relationship between a world coordinate system (p) and a sensor coordinate system (p '). The figure which shows the range point which shows a vehicle side surface. Graph of height curve created from object height data and frame. The flowchart which shows the process of extracting vehicle body side data and discriminating a parked vehicle. The figure which shows schematic structure of a shielding area extraction part. Depth curve graph created from extracted road surface point set. The flowchart which shows the process of extracting a shielding area and discriminating a parked vehicle. The figure which shows schematic structure of the parked vehicle detection system concerning one Embodiment of this invention. The figure which each shows the imaging | photography screen of a parked vehicle by a line scan camera, a characteristic locus | trajectory, and an analysis result. The figure which shows the state which binarized the divided image after dividing a spatiotemporal image. The figure explaining the relationship between the distance from the origin to the feature locus and the slope of the normal. The figure which shows a Hough space image. The figure which shows the process of setting a threshold value. The figure which shows the example which acquires an EPI image like 3rd Embodiment using a laser scanner. 3 is a table showing measurement conditions in Example 1. FIG. FIG. 6 is a diagram showing measurement results in Example 1. FIG. 6 is a table showing measurement conditions in Example 2. The figure which shows the measurement result in Example 2. FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Parked vehicle detection apparatus 2 ... Laser scanner 3 ... Parked vehicle 4 ... Road 5 ... Measurement vehicle 25 ... Data storage part 26 ... Program storage part 27 ... Imaging data storage part 28 ... Parked vehicle data storage part 29 ... Measurement condition storage part 32 ... Main program 33 ... Imaging data acquisition unit 34 ... Side surface data extraction unit 35 ... Shielded region extraction unit 36 ... Parked vehicle presence / absence determination unit 37 ... Number of parked vehicle number calculation unit 38 ... Parking region occupancy rate calculation unit 39 ... Speed / distance information Acquisition unit 40 ... processing result output unit 41 ... imaging data pre-processing unit 42 ... height calculation unit 43 ... height threshold setting unit 44 ... height data extraction unit 45 ... threshold update unit 46 ... side data output unit 47 ... reference frame Number setting unit 48 ... Imaging data pre-processing unit 50 ... Imaging data pre-processing unit 51 ... Depth calculation unit 52 ... Depth threshold setting unit 53 ... Depth data extraction unit 54 ... threshold update unit 55 ... shielded area output unit 56 ... reference frame number setting unit 60 ... parked vehicle detection device 61 ... line scan camera 62 ... road imaging unit 63 ... characteristic locus extraction unit 64 ... tilt calculation unit 65 ... distance calculation unit 66 ... Parking vehicle presence / absence discriminator 67 ... Feature trajectory

Claims (2)

Time and space components are synthesized in time series by combining multiple line images obtained by multiple line scans of an object on the road in the horizontal direction with a line scan camera mounted on a measurement vehicle traveling on the road. Obtaining an epipolar plane image (EPI image) having:
Extracting one or more straight lines connecting the positions of one or more feature points of the object in the line image in time series as one or more feature trajectories of the object in the EPI image;
Calculating an inclination of the extracted feature trajectory on coordinates having time and space components in the EPI image as coordinate axes ;
A step of calculating a depth distance to the feature point of the object according to the following formula based on the calculated inclination of the feature trajectory and the moving speed of the measurement vehicle ;

And a step of determining the presence or absence of the parked vehicle by comparing the calculated depth distance to the feature point with a predetermined threshold . An image data acquisition unit that acquires a plurality of line images obtained by scanning a target object on the road a plurality of times in a horizontal direction by a line scan camera mounted on a measurement vehicle traveling on the road;
A road imaging unit that acquires a horizontal epipolar plane image ( EPI image ) by synthesizing a plurality of line images of the object along a time series;
A feature trajectory extraction unit that extracts one or more straight lines connecting the positions of one or more feature points of the object in the line image in time series as one or more feature trajectories of the object in the EPI image;
An inclination calculation unit for calculating an inclination of the extracted feature locus on coordinates having time and space components in the EPI image as coordinate axes ;
A distance calculation unit that calculates the depth distance to the feature point of the target object according to the following formula based on the calculated inclination of the characteristic locus and the moving speed of the measurement vehicle ;

A parked vehicle detection system, comprising: a parked vehicle presence / absence determining unit that determines the presence / absence of a parked vehicle by comparing the calculated depth distance to a feature point with a predetermined threshold .

Images