JP2013214230A - Driving support device, driving support method, and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a driving support device capable of highly accurately determining presence/absence of a passersby such as pedestrians according to a traveling position of a vehicle.SOLUTION: A driving support device 1 includes: an image data acquisition unit 11; an overt passerby existence probability determination unit 13 that determines an overt passerby existence probability using a feature amount of the acquired image data; a position information acquisition unit 14 that acquires position information of vehicles; a potential passerby existence probability determination unit 17 that determines a potential passerby existence probability using the acquired position information; and a passerby existence determination unit 18 that determines presence/absence of a passersby in the periphery of the vehicle using the overt passerby existence probability and the potential passerby existence probability.

Description

  The present invention relates to a driving support device, a driving support method, and a computer program that support driving of a vehicle such as an automobile, and more particularly to a driving support device that supports driving by notifying a driver of the presence of a passerby around the vehicle. The present invention relates to a driving support method and a computer program.

  In recent years, as one of the driving support functions in vehicles such as automobiles, the presence or absence of passers-by such as pedestrians and bicycle users around the vehicle using a camera or a sensing device such as a laser radar provided in the vehicle, When the presence of a passer-by is detected, a passer-by detection function that informs the driver of that fact and calls attention is being adopted.

  In the detection of passers-by using a sensing device, there are false detections in which objects that are not pedestrians or bicycle users around the vehicle are detected as passers-by, and undetected problems that passers-by that should be detected are not detected as passers Can occur. Such false detection / non-detection may include sensing device performance such as resolution, disturbances in sensing such as occlusion (a part of the passerby in the back by an object in front) or thermal noise, Various factors, such as the accuracy of an algorithm when determining the presence or absence of a passerby using data acquired from a sensing device, are caused by a single time in some cases and overlapping in some cases.

  Even though there are people such as pedestrians and bicycle users in the vicinity of the vehicle, if there are many undetected cases, it has sufficient reliability as a driving support device for supporting safe driving I can't say that. In this case, it is possible to reduce the frequency of non-detection by increasing the detection sensitivity, that is, by setting a predetermined threshold value that is a criterion for determining whether or not the sensing result should be determined to be a passerby in advance. However, if the threshold value is simply lowered in this way, conversely, there is a possibility that erroneous detection frequently occurs in a situation where there are few pedestrians or the like. It is desirable for the driver to change the sensitivity of detection according to the situation such as the driving location and the driving date and time, because it is a great stress for the driver that unnecessary alerts are made due to erroneous detection.

  Therefore, when the traffic light state is blue or red, the density of searching for pedestrians from the vehicle periphery image is constant, while when the traffic light state is yellow, around the boundary between the sidewalk and the roadway. Some search density modes are switched so as to increase the density for searching for pedestrians (see Patent Document 1).

JP 2011-253214 A

  However, in the technique described in Patent Document 1, since the search density mode can be switched only at a place where there is a traffic light, the occurrence of undetected and false detection cannot be reduced at other places.

  The present invention has been made in view of the above-described problems, and provides a driving assistance device capable of accurately determining the presence or absence of a pedestrian or other passerby according to the traveling position of a vehicle. With the goal.

  The driving support device of the present invention uses a sensing data acquisition unit that acquires sensing data for detecting the presence of a passerby around a vehicle, and an explicit passer presence probability using the acquired feature value of the sensing data. An explicit passer existence probability determining unit for determining, a position information acquiring unit for acquiring vehicle position information, and a potential passer existence probability determining for determining a potential passer existence probability using the acquired position information And a passer presence determination unit that determines the presence or absence of passers around the vehicle using the explicit passer existence probability and the potential passer presence probability.

  According to this configuration, the presence / absence of a passerby is determined by using the actual passer existence probability determined using the sensing data and the potential passer existence probability determined based on the position information of the vehicle. Therefore, it can be determined that there is a passerby but does not exist, or it is determined that there is no passerby, and it is possible to reduce determination errors and accurately determine the presence or absence of a passerby.

  The driving support device of the present invention further includes a date information acquisition unit that acquires date information, and the potential passer existence probability determination unit further uses the acquired date information to detect the presence of the potential passer. The probability may be determined.

  According to this configuration, since the potential passer existence probability is determined in consideration of the date and time information in addition to the position information, it is possible to determine the potential passer existence probability in detail, and further reduce determination errors. be able to.

  The driving support device of the present invention further includes a weather information acquisition unit that acquires weather information, and the potential passer existence probability determination unit further uses the acquired weather information to detect the potential passer presence. The probability may be determined.

  According to this configuration, since the potential passer existence probability is determined in consideration of the weather information, the potential passer existence probability can be determined in detail, and further, determination errors can be reduced.

  In the driving support device of the present invention, the potential passer existence probability determining unit includes a potential passer existence probability map in which the potential passer existence probability calculated in advance is assigned to map data, and a map search unit. A probability value obtained by searching the potential passer existence probability map by the map search unit using the position information as a search key may be determined as the potential passer existence probability. .

  According to this configuration, when determining the potential passer existence probability, it is sufficient to search the potential passer existence probability map, and sensing is performed with a simple configuration without actually counting the passers. It is possible to compensate for detection errors caused by.

  In the driving support device of the present invention, the potential passer existence probability determining unit includes a potential passer existence probability map in which the potential passer existence probability calculated in advance is assigned to map data, and a map search unit. A probability value obtained by searching the potential passer existence probability map by the map search unit using the position information and the date / time information as a search key is determined as the potential passer existence probability. There may be.

  According to this configuration, when determining the potential passer existence probability, the date and time information is also used as a search key in a detailed manner, although it is a simple configuration of searching for a potential passer presence probability map, A detection error due to sensing can be supplemented.

  In the driving support device of the present invention, the potential passer existence probability determining unit includes a potential passer existence probability map in which the potential passer existence probability calculated in advance is assigned to map data, and a map search unit. A probability value obtained by searching the potential passer existence probability map by the map search unit using the position information and the weather information as search keys is determined as the potential passer existence probability. There may be.

  According to this configuration, when determining the potential passer existence probability, the weather information is also used as a search key in a fine structure, although it is a simple configuration of searching for a potential passer presence probability map, A detection error due to sensing can be supplemented.

  In the driving assistance device of the present invention, the potential passer existence probability may be a probability value calculated by calculating a plurality of different numerical values according to information on at least one of position, time, and weather. .

  According to this configuration, the potential passer existence probability can be changed depending on, for example, the position or time of day, the day of the week, the season, the holiday, or whether it is raining. It is possible to reflect the driving situation such as the vehicle travel position, date and time, and weather in the passer-by presence determination using the existence probability.

  In the driving support device of the present invention, at least one of the plurality of numerical values may be a numerical value determined based on demographic data.

  According to this configuration, for example, the number of passers estimated based on demographic data can be corrected using other numerical values. Can be determined.

  In the driving support device of the present invention, at least one of the plurality of numerical values is set so that the potential passer existence probability is high within a range having a predetermined distance from a predetermined facility or a predetermined road section. It may be a numerical value.

  According to this configuration, for example, in the vicinity of a facility where the presence of a passerby is predicted, a numerical value that increases the probability of potential passer presence is set as a coefficient value, thereby affecting the presence of the passerby. Possible factors can be reflected in the probability of potential passers.

  In the driving support device of the present invention, at least one of the plurality of numerical values is a numerical value set so that the potential passer existence probability is high within a range having a predetermined distance from a traffic accident occurrence place. May be.

  According to this configuration, the coefficient value can be set to a numerical value that increases the probability of potential passers in places where traffic accidents are likely to occur, and there are factors that directly affect the number of passers. Even where there is no place, it is possible to increase the probability of potential passers-by in a place where attention is required for driving, and to alert the driver.

  In the driving support device of the present invention, at least one of the plurality of numerical values may be a numerical value set according to any of a time zone, a day of the week, and a month.

  According to this configuration, for example, a numerical value that increases or decreases the potential passer existence probability for each time zone, day of the week, or season can be set as a coefficient value, and it is fine and reliable. A high potential passer presence probability can be determined.

  In the driving support device of the present invention, at least one of the plurality of numerical values may be a numerical value set according to event information to be held.

  According to this configuration, it is possible to set, as a coefficient value, a numerical value that increases the probability of potential passer presence according to the event that is expected to have a large number of passers-by, and is detailed and highly reliable. A potential passer presence probability can be determined.

  In the driving assistance device of the present invention, the passer presence determination unit uses the explicit passer existence probability and the potential passer presence probability to calculate a passer presence probability and 1 to the passer presence probability. Calculating a passer absence probability that is a subtracted value, and determining that a passer is present when a ratio of the passer presence probability and the passer absence probability exceeds a predetermined threshold; Also good.

  According to this configuration, for example, it is possible to more accurately determine the presence of a passer compared to a case where the passer existence probability and the passer non-existence probability are simply compared to determine the presence of the passer. .

  In the driving assistance device of the present invention, the predetermined threshold value may be a ratio of a loss value when it is erroneously determined that there is no passer and a loss value when it is erroneously determined that a passer-by exists. Good.

  According to this configuration, it is possible to minimize a negative effect (cost) due to a determination error when it is determined that there is a passerby but does not exist and when it is determined that there is no passerby.

  The driving support method of the present invention includes a step of acquiring sensing data for detecting the presence of a passerby around a vehicle, and a step of determining an apparent passer presence probability using a feature amount of the acquired sensing data. A step of acquiring vehicle position information, a step of determining a potential passer existence probability using the acquired position information, and the explicit passer presence probability and the potential passer presence probability. And determining whether or not there are passersby around the vehicle.

  According to this configuration, the presence / absence of a passerby is determined by using the actual passer existence probability determined using the sensing data and the potential passer existence probability determined based on the position information of the vehicle. Therefore, it can be determined that there is a passerby but does not exist, or it is determined that there is no passerby, and it is possible to reduce determination errors and accurately determine the presence or absence of a passerby.

  Yet another embodiment of the present invention is a computer program for causing a computer to execute the driving support method described above.

  According to the present invention, it is possible to accurately determine the presence or absence of a passerby such as a pedestrian according to the traveling position of the vehicle.

The block diagram which shows the structure of the driving assistance apparatus in embodiment of this invention. The flowchart which shows operation | movement of the driving assistance apparatus in embodiment of this invention The flowchart which shows the flow of an explicit passer existence probability determination process in embodiment of this invention The block diagram which shows the structure of the potential passer existence probability determination part in embodiment of this invention The flowchart which shows the flow of the production | generation process of the potential passer presence probability map in embodiment of this invention

  Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration of the driving support apparatus according to the present embodiment. The driving support device 1 includes an image data acquisition unit 11, a feature amount extraction unit 12, an explicit passer existence probability determination unit 13, a position information acquisition unit 14, a date / time information acquisition unit 15, a weather information acquisition unit 16, and a potential passerby. A presence probability determination unit 17, a passer-by presence determination unit 18, and a notification control unit 19 are provided.

  The vehicle in which the driving support device 1 is used is provided with a camera 20, a GPS receiver 30, a clock unit 40, a wiper control unit 50, and a notification unit 60.

  The camera 20 is an imaging device such as a small CCD camera or a CMOS camera installed in the vehicle in order to photograph the situation around the vehicle including a passerby. The camera 20 may be installed near the upper portion of the windshield and photograph the front of the vehicle, or may be provided at the rear of the vehicle and photograph the rear periphery.

  The image data acquisition unit 11 acquires image data around the vehicle captured by the camera 20 and outputs the acquired image data to the feature amount extraction unit 12.

  The feature amount extraction unit 12 extracts a feature amount for identifying a passer-by from the image data acquired by the image data acquisition unit 11 and outputs the feature amount to the explicit passer-existence probability calculation unit 13. The feature amount extracted here is data that can be compared with predetermined index data indicating a pattern of a passerby that is a detection target. The passersby who are subject to the presence determination of the driving support device 1 include pedestrians walking in various orientations and postures with respect to the vehicle and those riding on bicycles. Therefore, the feature amount here is preferably information that can detect the presence of a passerby regardless of such diversity. In the present embodiment, HOG (Histograms of Oriented Gradients) feature values are extracted.

  The explicit passer existence probability determination unit 13 performs pattern recognition using the feature amount data output from the feature amount extraction unit 12, and the probability that the passerby is included in the image data around the vehicle captured by the camera 20. Is output to the passer-by presence determination unit 18. Here, the actual passer existence probability means the probability that a passerby exists when a feature value is given, and it relates to the surrounding situation of the vehicle actually observed and measured using a sensor such as a camera. It is directly calculated from the feature amount of information. In the present embodiment, the actual passer existence probability is calculated by pattern recognition using a neural network. Therefore, in the obvious passer existence probability determining unit 13, advance learning is performed on the feature amount of the passer.

  The position information acquisition unit 14 acquires position information indicating the current position of the vehicle from the GPS receiver 30 mounted on the vehicle, and outputs the position information to the potential passer presence probability determination unit 17. The position information acquisition unit 14 acquires position information when the surroundings of the vehicle are captured by the camera 20 or when image data is acquired by the image data acquisition unit 11. That is, the acquisition of position information may be performed at the same time or substantially at the same time as the shooting or acquisition of image data, but does not have to be performed at the same time or substantially at the same time. The position information may be acquired within a certain period of time.

  The date / time information acquisition unit 15 acquires date / time information from the clock unit 40 mounted on the vehicle and outputs the date / time information to the potential passer presence probability determination unit 17.

  The weather information acquisition unit 16 acquires information regarding the operation of the wiper as weather information from the wiper control unit 50 that controls the wiper attached to the vehicle, and outputs the information to the potential passer presence probability determination unit 17. That is, in the weather acquisition unit 16, information regarding the operation of the wiper is associated with the weather information. For example, if the wiper is intermittent, “light rain”, if it is continuously operating at a constant pace, “rain”, if it is continuously operating at a fast pace, “heavy rain”, etc. It is output to the existence probability determining unit 17. Note that only information regarding whether or not the wiper is operating may be acquired and output to the potential passer existence probability determining unit 17.

  The potential passer existence probability determination unit 17 is based on the position information output from the position information acquisition unit 14, the date information output from the date information acquisition unit 15, and the weather information acquired from the weather information acquisition unit 16. Determine the potential passer presence probability. As will be described in detail later, the potential passer existence probability determining unit 17 includes a potential passer existence probability map. Based on the position information, the date and time information, and the weather information, the potential passer existence probability is provided. Search the map. In the potential passer existence probability map, the passer existence probability calculated using the position X and the date and time T as parameters is represented on the map data. That is, the potential passer existence probability refers to a probability that a passerby exists when a certain position X and time T are given.

  Neither the position information nor the date / time information can be thought of from the information itself as a feature quantity for detecting a person, and therefore the presence of a passerby cannot be detected only by such information. However, it is possible to make a prediction based on statistical data regarding the presence of a passerby at a specific date and a specific location. Similarly, weather information can also be used for such predictions as a factor that affects the presence of passers-by. Position information, date and time information as parameters are acquired, but in order to estimate the existence of a passerby without using sensing data such as image data obtained by actually sensing the situation around the vehicle, explicit traffic In the sense that passers who are not considered when calculating the presence probability of a person may be included, the presence probability of a passerby determined based on location information, date and time information, and weather information is “potential” it can.

  The passer-by presence determination unit 18 uses the explicit passer-existence probability output from the explicit passer-existence probability determination unit 13 and the potential passer-existence probability output from the potential passer-existence probability determination unit 17. The presence or absence of passersby is determined. In order to minimize the risk of two judgment mistakes (misjudgment) when it is determined that there is no passer by mistake, and when it is determined that there is a passer by mistake, In this embodiment, it follows Bayesian decision theory. That is, a low cost is selected from the negative effect (cost) caused by the determination error of the former and the cost generated by the determination error of the latter. Specifically, if there is a passer if the ratio of the passer existence probability, that is, the probability that a passer exists, and the passer non-existence probability, that is, the probability that no passer exists, exceeds a predetermined threshold. judge. The passer existence probability is given as being proportional to the product of the explicit passer existence probability and the potential passer existence probability. On the other hand, the passer non-existence probability is given as being proportional to the product of (1-apparent passer existence probability) and (1-potential passer existence probability).

  While the actual passer existence probability can give a detection result more suitable to the actual situation of the surroundings of the vehicle, a detection error is likely to occur due to disturbance during sensing. On the other hand, although the potential passer existence probability does not necessarily reflect the actual situation of the surroundings of the vehicle, it provides a reasonable guess based on the traveling position, date and time of the vehicle. In this embodiment, by determining the presence or absence of a passer using both the explicit passer existence probability and the potential passer existence probability, the erroneous determination is reduced, and the presence or absence of the passer is accurately determined. Can be determined.

  As described above, the position information acquisition by the position information acquisition unit 14 is performed when the periphery of the vehicle is photographed by the camera 20 or when the image data acquisition unit 11 acquires image data. The passer existence probability and the potential passer existence probability indicate the existence probabilities of passers at substantially the same position or positions close to each other. Accordingly, the passer-by presence determination unit 18 can determine the presence / absence of a passer-by at the position using the two passer-existence probabilities. On the other hand, the acquisition of the date / time information from the clock unit 40 by the date / time information acquisition unit 15 may be performed every time shooting by the camera 20 or acquisition of image data is performed, and after the date / time information is acquired once, the time zone is switched. The date and time information may be stored, and the stored date and time information may be output to the potential passer existence probability determining unit 17. Further, the weather information acquisition by the weather information acquisition unit 16 may be performed every time shooting by the camera 20 or acquisition of image data, or may be performed according to the start or end of the wiper operation, operation switching, or the like.

  When the passer presence determination unit 18 determines that a passer-by exists, the notification control unit 19 instructs the notification unit 60 to notify the driver of the presence of the passer-by in order to alert the driver. Output. The notification unit 60 is, for example, a display or a speaker mounted on the vehicle, and displays a message indicating that a passerby is present or outputs a similar voice message.

  The driving support device 1 stores a program for realizing each of these functions.

  Next, operation | movement of the driving assistance apparatus 1 provided with such a structure is demonstrated using the flowchart of FIG.

  First, the obvious passer existence probability is determined (step S21). As described above, the actual passer presence probability is calculated based on the feature amount extracted from the image data. The calculated actual passer existence probability can be expressed as p (α1 | V) using the given feature value V. Here, α1 represents a state where a passerby exists.

  Next, the potential passer presence probability is determined (step S22). As described above, the potential passer existence probability is determined by searching the potential passer existence probability map using the position information, the date and time information, and the weather information. The potential passer existence probability can be expressed as p (α1 | X, T) using the place X and the time T.

Further, a passer presence probability and a passer absence probability are calculated (step S23). The passer existence probability p1 can be obtained by the expression (1) using the explicit passer existence probability and the potential passer existence probability.
p1 = rp (α1 | V) p (α1 | X, T) / p (α1) (1)
Where p (α1) is
p (α1) = Σ _{X, T} p (α1 | X, T) / Σ _{X, T} 1
It shows the probability that a passerby exists at any date, time and place.
On the other hand, the passer non-existence probability p2 can be obtained by Expression (2) using the obvious passer existence probability and the potential passer existence probability.
p2 = r (1-p (α1 | V)) (1-p (α1 | X, T)) / (1-p (α1)) (2)

In the formulas (1) and (2), r is
r = p (V) is a coefficient given by p (X, T) / p (V, X, T), which is canceled by the left side and the right side of the inequality in the threshold processing described later. There is no need to find a numerical value.

  It is determined whether or not there is a passer by using the passer existence probability p1 and the passer non-existence probability p2 thus obtained (step S24). At this time, by simply comparing p1 and p2, it may be determined that a passerby exists when p2 <p1, but in this embodiment, a negative effect (cost) due to erroneous determination is obtained. In order to minimize, threshold processing is performed according to Bayesian decision theory.

  In other words, if c1 is the loss value when it is erroneously determined that there is no passer and c2 is the loss value when it is determined that there is a passer by mistake, then there is a passer if c2p2 <c1p1. Judge that it is. c1 and c2 are, for example, c2 = 10c1, c1 = 1, etc. when the cost when it is erroneously determined that there is a passer is 10 times the cost when it is erroneously determined that there is no passer I can leave. The ratio between c1 and c2 needs to be determined in advance, but such threshold processing minimizes the cost due to misjudgment under any position (X) and date and time (T).

  In step S24, when it is determined that there is a passerby, the notification control unit 19 outputs a command to notify the notification unit 60 of the fact, and the notification unit 60 notifies the notification by a predetermined method. This is performed (step S25).

  FIG. 3 is a flowchart showing the flow of the process of determining the obvious passer existence probability in step S21 of FIG. First, the image data acquisition unit 11 acquires image data around the vehicle photographed by the camera 20 (step S31).

  Subsequently, HOG feature values for identifying passersby are extracted from the acquired image data (step S32). The HOG feature amount is a feature amount that can represent a rough shape of an object, and is robust to a slight change in shape, and is often used for human detection. The feature amount extraction unit 12 divides the acquired image data into local regions (cells), generates a gradient of luminance gradient strength and gradient direction for each cell, and generates a block unit including a plurality of cells. The histogram is normalized to obtain the HOG feature amount.

  Further, the explicit passer existence probability determining unit 13 inputs the extracted HOG feature quantity to the neural network, and determines the explicit passer existence probability (step S33). For this reason, prior to the determination of the actual passer existence probability, it is necessary to perform prior learning on the neural network. Specifically, the neural network is trained using the back propagation method so that each target variable approaches 1 and 0 from the feature amount (positive example) of the passerby and the other feature amount (negative example). The coupling coefficient between each layer of the network is adjusted repeatedly. The output of the neural network generated in this way can be regarded as the existence probability of the passer-by in the state where the feature amount V is given, that is, the obvious passer-existence probability (p (α1 | V)). .

  FIG. 4 is a block diagram showing a detailed configuration of the potential passer existence probability determining unit 17. As shown in FIG. 4, the potential passer existence probability determining unit 17 includes a potential passer existence probability map search unit 161, a potential passer existence probability map 162, a demographic database 163, a traffic accident statistical database 164, a map, and the like. An information database 165, an event information database 166, a data reading unit 167, an estimated passer-by number calculation unit 168, and a probability conversion unit 169 are provided.

  The potential passer existence probability map search unit 161 uses the vehicle position information acquired by the position information acquisition unit 14, the date information acquired by the date information acquisition unit 15, and the weather information acquired by the weather information acquisition unit 16. The potential passer existence probability map 162 is searched as a search key, and the obtained potential passer existence probability is output to the passer presence determination unit 18.

  The potential passer existence probability map 162 represents the probability that a passerby exists at a certain position (X) at a certain date and time (T) on the map. As will be described later, the demographic database 163, traffic It is generated based on information stored in the accident statistics database 164, the map information database 165, and the event information database 166.

  The demographic database 163 stores data on the population and population density for each region, for example, a district or a chome unit that is a division of a municipality or smaller, and data on daytime population for each region. The traffic accident statistics database 164 stores data on the number and location of traffic accidents occurring in a certain period for each region. The map information database 165 stores general map information used for car navigation devices and the like, and includes information related to roads such as business hours and school routes such as stores and public facilities. The event information database 166 stores event information including information on the date and place of the event. This event information is acquired from time to time using, for example, a predetermined communication means from a server that distributes the event information.

  Note that the demographic database 163, the traffic accident statistics database 164, and the event information database 166 may hold numerical data mapped on a map, or the numerical data and the position or location corresponding to the numerical data. The range may be specified and held by latitude and longitude.

  The data reading unit 167 reads data stored in the demographic statistics database 163, the traffic accident statistics database 164, the map information database 165, and the event information database 166, and outputs the data to the estimated passer-by number calculation unit 168.

  The estimated number of passers-by calculation unit 168 calculates the estimated number of passers-by at the place X and date / time T based on the data read from each database, and outputs it to the probability conversion unit 169.

  The probability conversion unit 169 converts the estimated number of passers-by into 0-1 probability values. The potential passer existence probability map 162 represents the estimated passer existence probability calculated by the probability conversion unit 169 on the map data.

  FIG. 5 is a flowchart showing a flow of processing for generating the potential passer existence probability map 162. In the present embodiment, the potential passer existence probability map 162 prior to the acquisition of the position information by the position information acquisition unit 14, the acquisition of the date information by the date information acquisition unit 15, and the weather information acquisition by the weather information acquisition unit 16. Is generated.

  First, the data reading unit 167 reads data relating to population density for each region from the demographic database 163, and calculates the basic estimated number of passers-by (step S511). Here, the basic estimated number of passers is the estimated number of passers-by at a predetermined time within a unit area area belonging to the area, which is estimated from the population density of each area. Specifically, for example, it is assumed that the population density (a person / square kilometer) of the A ward is read from the demographic database, and 1% of the residents are walking in the neighborhood at 3 pm on a typical weekday. In this case, the basic estimated number of passers-by is obtained by a × 0.01, and it is estimated that 0.01a passers-by exist in each area of 1 square kilometer belonging to the A ward.

  Next, within a range of a predetermined distance or less from a predetermined facility or road section included in the map data of the map information database 165, the coefficient value determined for each attribute of the facility or road section is multiplied by the basic estimated number of passersby. (Step S512). The predetermined facility is a public facility such as a station, a bus stop, a store, or a library, for example, and the predetermined road section is, for example, a part or all of a road including a school road or an intersection. In general, there is a high possibility that there are many passersby in such facilities or road sections. Therefore, in this step, a coefficient value larger than 1 is multiplied by the basic estimated number of passers-by. The specific coefficient value is, for example, 3.0 if the road section attribute is an intersection, and 5.0 if the facility attribute is a station.

  Further, the location range where the coefficient values are multiplied, that is, the distance from a predetermined facility or road section may be different for each facility attribute or road section attribute. For example, if the road section attribute is an intersection, the distance is within 30 meters. If the facility attribute is a station, the coefficient value is multiplied by 5.0 at a distance within 100 meters. In this way, by setting a coefficient value that increases the estimated number of passers-by around facilities or road sections where the presence of passers-by is expected, it will not be included in demographic data, but passers-by It is possible to reflect factors that can affect

  Furthermore, using the read map data, the coefficient value corresponding to the road attribute, that is, the type of road is multiplied by the value of step S512 (step S513). Specifically, since it is estimated that there are few passers-by on an expressway, the coefficient value is multiplied by 0.1, and if it is a general road, the coefficient value is multiplied by 1.0.

  Next, the coefficient value determined from the accident occurrence number data read from the accident statistics database 164 is multiplied by the value in step 513 (step S514). For example, the coefficient value is multiplied by 2.0 in a range within 30 meters of a point where the average number of accidents per year is one or more. As a result, even in places where there are no factors that directly affect the number of passers-by, there is a high probability of traffic accidents, and in places where attention is required to drive, the probability of potential passers-by is increased and the driver is alerted. be able to.

  In step S515, for a predetermined road section attribute included in the map data, the value calculated in step S514 is further multiplied by a predetermined coefficient value. Specifically, for example, when the road section attribute is a narrow alley, an intersection with poor visibility, or a multi-difference road, the coefficient value 2.0 is set. The road section attribute in step S512 predicts the presence of many passers-by, while the road section attribute to be multiplied by the coefficient value in this step generally has a high probability of occurrence of an accident. is there. Therefore, similarly to the coefficient value calculation in step S514, the potential passer existence probability can be increased in a place where attention is required for driving, and the driver can be alerted.

  Then, the coefficient value based on the daytime population data read from the demographic database 163 is multiplied by the value in step S515 (step S516). The coefficient value at this time is given by, for example, daytime population / resident population. By multiplying such coefficient values, the size of the city can be reflected in the estimated number of passers-by in an area where the number of people in the daytime is larger than the number of inhabitants as in the city. When the daytime population is unknown, the coefficient may be uniformly set to 3.0.

  Through the processing up to step S516 described above, an estimated passer number map to which the estimated passer number at a certain date and time is assigned can be generated. From this map, probability value conversion can be performed to generate a potential passer existence probability map 162. However, even at the same position, a significant difference can occur in the number of passers estimated by the time zone. Therefore, in the present embodiment, the estimated number of passers-by map generated by the processing up to step S516 is prepared for each of a plurality of time segments such as every time zone and every day of the week by the following steps.

  Specifically, in step S517, the coefficient value determined for each day, that is, for each time period obtained by dividing 24 hours by 1 hour, is multiplied by the numerical value obtained in step S516. For example, from 7:00 am to 8:00 am, as the commuting time zone, it is estimated that there are more passers than the number of passers at 3:00 pm assumed in step S511. At the same time, the time zone from noon to 1 pm is not the same as the commute time zone as a lunch break time zone, but there are more outsiders than 3 pm, so the coefficient value is 2.0. Multiply The coefficient value to be multiplied in this step S517 may be set according to the business hours of the store or facility included in the map data. Therefore, even in the same time zone, a different coefficient value is set for each area. Also good. This is calculated for all 24 time zones, and the process proceeds to step S518. The coefficient value may be set by further dividing the time zone.

  In the subsequent step S518, coefficient values set by paying attention to a time segment larger than that in step S517, such as a day of the week in a week, a date such as a holiday in a year (month day), or a season (month), are set to the numerical values in step S517. Multiply each. For example, 1.0 on weekdays (Monday to Friday), 1.5 on Saturdays and Sundays when more passers are expected, and winter (December to February) Multiply by 0.8, and multiply by 1.2 for other seasons. In areas such as office districts where there are fewer passers on holidays, even if the time interval is the same, such as setting the coefficient value for weekdays higher than the coefficient value for holidays, there are even different coefficient values for each area. It may be set.

  Subsequently, a coefficient value corresponding to the weather is multiplied (step S519). The number and specific numerical values may vary depending on what information the weather information acquisition unit 16 acquires as weather information. For example, according to the wiper operation, intermittent operation is acquired as “light rain”, normal continuous operation as “rain”, and continuous operation at a fast pace as “heavy rain” as weather information. 0.9, 0.7 in case of rain, 0.3 in case of heavy rain, 1 in other cases.

  Furthermore, when a large event or the like is held, it is estimated that there are many passersby in the vicinity of the holding place and in a time zone close to the start time and the end time. Therefore, in step S520, a predetermined coefficient is multiplied by the numerical value in step S519 in accordance with the place and date of the event to be held. Specifically, the data reading unit 167 reads information on the event holding facility and the holding date / time from the event information database 166, and the estimated number of passersby determination unit 168 includes a plurality of data before and after the start time on the day when the event is held. In the time zone and a plurality of time zones before and after the end time, the coefficient value 10.0 is multiplied by the numerical value in step S519 for the area within the predetermined distance range from the facility. This coefficient value may be different for each facility where the event is held.

  The numerical value obtained by sequentially multiplying the basic estimated number of passers by the coefficient value by the processing up to step S520 described above is the estimated number of passers n (X, T) at a certain position X date T.

In step S521, the estimated number of passers-by is converted into a probability value. In the present embodiment, the real value in step S520 is converted into a value between (0, 1) using a sigmoid function, and is made probabilistic. This probability value is the potential passer existence probability P _c (X, T). P _c (X, T) is obtained by the following equation.
P _c (X, T) = σa (n (X, T) −n _m )
Here, _nm is an average value of n. Σa (x) is a sigmoid function having a gain a,
σa (x) = 1 / (1 + exp (−ax)). In order to prevent the deviation of the probability value, the gain a uses a function std (x) that returns the standard deviation of x,
a = 1 / std (n). A potential passer existence probability map 162 represents the probability values obtained in this way on map data. As described above, the potential passer existence probability map 162 is used for determining the potential passer existence probability using the position information, the date information, and the like.

  As described above, according to the driving support apparatus 1 in the present embodiment, the actual passer existence probability calculated by extracting the feature amount from the image around the vehicle captured by the camera 20 and the GPS receiver 30. The potential passerby determined by searching the potential passer existence probability map 162 using the acquired position information, date and time information acquired from the clock unit 40, and weather information acquired from the wiper control unit 50. The existence probability is used to determine the presence or absence of a passerby around the vehicle, so depending on the location of the vehicle, it can be determined that there is a passerby but does not exist, or there is no passerby. Judgment mistakes can be reduced.

  The embodiments of the present invention have been described above by way of example, but the scope of the present invention is not limited to these embodiments, and can be changed or modified according to the purpose within the scope of the claims. is there.

  For example, in the above description, a camera is used as a sensor and image data is used as sensing data. However, laser radar may be used as a sensor, and scanning data may be used as sensing data.

  Moreover, although the above description demonstrated the case where the weather information acquisition part 16 acquires weather information from the wiper control part 50, it acquires weather information from a thermometer instead of the wiper control part 50 or with it. In this case, the potential passer existence probability may be calculated by multiplying a coefficient value set according to the temperature category. In addition, weather information such as temperature and weather may be acquired from a television, radio, network service, or the like.

  In the above description, the case where the notification control unit 19 notifies the presence of the passer by the notification control unit 19 when the passer presence determination unit 18 determines that the passer is present has been described. Alternatively, driving of the vehicle may be supported by brake control and speed control of the vehicle on which the driving support device 1 is mounted together with the notification. That is, the driving support device 1 may be configured not to include the notification control unit 19.

  In the above description, the case where the potential passer existence probability map 162 is obtained by assigning the potential passer existence probability on the map data has been described. It may be associated with area data, and in determining the potential passer existence probability, an area including a position indicated by the acquired position information, a position close thereto, or a position indicated by the acquired position information Alternatively, the potential passer existence probability associated with an area in the vicinity thereof may be output as a search result.

  Further, in the above description, the estimated number of passers-by at a given date and time is assigned to the estimated passer-by-number map on the map data, according to the coefficient value according to the date and time information such as every time zone, every day of the week, etc. In the above description, the potential passer existence probability map 162 is generated by multiplying the coefficient value corresponding to the event information and the event value according to the event information, but at least one of the date / time information, the weather information, and the event information is not used. The potential passer presence probability map 162 may be generated.

  In the above description, the estimated number of passers and the potential passer existence probability are calculated by sequentially multiplying the basic estimated passers determined based on the data stored in the demographic database 163 by the coefficient value. As described above, the basic estimated number of passers-by may be determined based on other data, or the probability value is first determined based on data stored in a certain database without determining the basic estimated number of passers-by. , And the probability value of the potential passer may be calculated by repeatedly adjusting the probability value with a plurality of other data.

  Furthermore, in the above description, the case where the potential passer existence probability determining unit 17 includes the potential passer existence probability map 162 in advance has been described. Or after acquiring date / time information and the like, the probability of potential passers-by may be calculated based on the demographic database 163 or the like.

  The present invention has an effect of being able to accurately determine the presence or absence of a pedestrian or the like of a pedestrian according to the traveling position of the vehicle, and as a driving support device that supports driving of a vehicle such as an automobile Useful.

DESCRIPTION OF SYMBOLS 1 Driving assistance apparatus 11 Image data acquisition part 12 Feature-value extraction part 13 Prominent passer existence probability determination part 14 Location information acquisition part 15 Date information acquisition part 16 Weather information acquisition part 17 Potential passer existence probability determination part 18 Passer Presence determination unit 19 Notification control unit 20 Camera 30 GPS receiver 40 Clock unit 50 Wiper control unit 60 Notification unit 161 Potential passer existence probability map search unit 162 Potential passer existence probability map 163 Demographic database 164 Traffic accident statistics database 165 Map information database 166 Event information database 167 Data reading unit 168 Estimated number of passers 169 Probability conversion unit

Claims (16)

A sensing data acquisition unit for acquiring sensing data for detecting the presence of a passerby around the vehicle;
An explicit passer existence probability determining unit for determining an explicit passer existence probability using the acquired feature amount of the sensing data;
A position information acquisition unit for acquiring position information of the vehicle;
A potential passer existence probability determining unit that determines a potential passer existence probability using the acquired position information;
A passer presence determination unit for determining presence or absence of passers around the vehicle using the explicit passer existence probability and the potential passer presence probability;
A driving support apparatus comprising: A date and time information acquisition unit for acquiring date and time information;
The driving support device according to claim 1, wherein the potential passer existence probability determining unit further determines the potential passer existence probability using the acquired date and time information. Further equipped with a weather information acquisition unit for acquiring weather information,
The driving support device according to claim 1, wherein the potential passer existence probability determining unit further determines the potential passer existence probability using the acquired weather information. The potential passer existence probability determining unit includes:
A potential passer existence probability map in which the potential passer existence probability calculated in advance is assigned to the map data, and a map search unit;
2. The probability value obtained by searching the potential passer existence probability map by the map search unit using the position information as a search key is determined as the potential passer existence probability. The driving support device according to 1. The potential passer existence probability determining unit includes:
A potential passer existence probability map in which the potential passer existence probability calculated in advance is assigned to the map data, and a map search unit;
Using the position information and the date / time information as a search key, the probability value obtained by searching the potential passer existence probability map by the map search unit is determined as the potential passer existence probability. The driving support device according to claim 2. The potential passer existence probability determining unit includes:
A potential passer existence probability map in which the potential passer existence probability calculated in advance is assigned to the map data, and a map search unit;
A probability value obtained by searching the potential passer existence probability map by the map search unit using the position information and the weather information as a search key is determined as the potential passer existence probability. The driving support device according to claim 3.   The potential passer existence probability is a probability value calculated by calculating a plurality of different numerical values according to information on at least one of position, time, and weather. The driving support device according to any one of the above.   The driving support device according to claim 7, wherein at least one of the plurality of numerical values is a numerical value determined based on demographic data.   At least one of the plurality of numerical values is a numerical value set so that the potential passer existence probability is high within a range having a predetermined distance from a predetermined facility or a predetermined road section. The driving support device according to claim 7.   8. At least one of the plurality of numerical values is a numerical value set so that the potential passer existence probability is high within a range having a predetermined distance from a traffic accident occurrence place. The driving support device according to 1.   The driving support device according to claim 7, wherein at least one of the plurality of numerical values is a numerical value set according to any of a time zone, a day of the week, and a month.   The driving support apparatus according to claim 7, wherein at least one of the plurality of numerical values is a numerical value set according to event information to be held. The passer-by presence determination unit
Using the explicit passer existence probability and the potential passer existence probability, a passer existence probability and a passer non-existence probability that is a value obtained by subtracting the passer existence probability from 1 are calculated.
The driving support device according to any one of claims 1 to 12, wherein a passer is determined to be present when a ratio between the passer existence probability and the passer absence probability exceeds a predetermined threshold. .   The predetermined threshold is a ratio of a loss value when it is erroneously determined that no passer is present and a loss value when it is erroneously determined that a passer is present. Driving assistance device. Obtaining sensing data for detecting the presence of passersby around the vehicle;
Determining an apparent passer presence probability using the acquired feature value of the sensing data;
Obtaining vehicle position information;
Determining a potential passer presence probability using the acquired location information;
Determining the presence or absence of a passerby around the vehicle using the explicit passer presence probability and the potential passer presence probability;
A driving support method comprising:   A computer program for causing a computer to execute the driving support method according to claim 15.

Images