JP5622766B2 - Route search device - Google Patents

Description

  The present invention relates to a route search device that generates a route having a low encounter probability with a pedestrian.

  Conventionally, driving assistance devices for supporting safe driving have been proposed. For example, in Patent Document 1, a pedestrian within a predetermined area is detected from an image acquired from a pedestrian detection camera, and the front of the host vehicle is detected based on the detected pedestrian state of the pedestrian and the traveling state of the host vehicle. A vehicle driving support device is disclosed in which a dangerous area is set in the vehicle, the degree of danger is determined depending on whether the pedestrian is in the dangerous area, and the driver is alerted.

JP 2009-99856 A

  However, with the technique disclosed in Patent Document 1 described above, encounters with pedestrians cannot be avoided, and there is a problem that the stress experienced by the driver due to encounters with pedestrians cannot be reduced.

  The present invention has been made to solve the above-described problems, and an object thereof is to provide a route search device that preferentially searches a route with a low probability of encountering a pedestrian and sets the route as a travel route. To do.

The route search device according to the present invention integrates a human detection unit that detects a pedestrian, a pedestrian information detected by the human detection unit, and own vehicle position information from a captured image obtained by photographing a predetermined area from the own vehicle. A data integration unit that generates the integrated data, a data storage unit that stores the integrated data generated by the data integration unit, an operation input unit that accepts destination setting and route generation condition specification, and an operation input unit When a route generation condition with a low encounter probability with a pedestrian is specified , a route search unit that searches for a route from the current position of the vehicle to the set destination based on the specification is stored in the data storage unit. Based on the encounter probability calculated by the encounter probability calculation unit and the encounter probability calculation unit that calculates the encounter probability of the constituent route that constitutes the route searched by the route search unit using the integrated data and the vehicle position information Configuration path A weight setting unit for setting the weight, and the route generating section that refers to the weight of the structure path weight setting unit has set, it selects a path having a smallest weight among the searched route as a travel route, and the captured image, human An attribute estimation unit that estimates a pedestrian attribute from the pedestrian information detected by the detection unit, and the data integration unit integrates the pedestrian attribute information estimated by the attribute estimation unit and the vehicle position information. Integrated data is generated, the encounter probability calculation unit calculates the attribute-specific encounter probability for each route for each pedestrian attribute information, and the calculated attribute-specific encounter probability is preset for each attribute of the pedestrian The encounter probability is calculated by applying the weight .

  According to the present invention, it is possible to accumulate pedestrian information on a road that is running, preferentially search a route with a low probability of encountering a pedestrian based on the accumulated pedestrian information, and set it as a traveling route. Can do. Thereby, the stress which a driver | operator receives by encounter with a pedestrian can be reduced. Furthermore, the occurrence of a contact accident with a pedestrian can be prevented.

1 is a block diagram illustrating a configuration of a route search device according to a first embodiment. FIG. 7 is an explanatory diagram illustrating a processing operation of the route search device according to the first embodiment. 4 is a flowchart showing a data collection processing operation of the route search device according to the first embodiment. 4 is a flowchart illustrating a route search processing operation of the route search device according to the first embodiment. 5 is a flowchart showing a route generation operation of the route search device according to the first embodiment. It is the figure which showed map information in the graph. 6 is an explanatory diagram illustrating an example of setting an optimum route of the route search device according to the first embodiment. FIG. It is a block diagram which shows the structure of the route search apparatus by Embodiment 2. FIG. 10 is an explanatory diagram illustrating a processing operation of the route search device according to the second embodiment. 6 is a flowchart showing a data collection processing operation of the route search apparatus according to the second embodiment.

Embodiment 1 FIG.
First, the configuration and processing operation of the route search apparatus according to the first embodiment will be described with reference to FIG. 1 and FIG. FIG. 1 is a block diagram showing the configuration of the route search apparatus according to the first embodiment of the present invention, and FIG. 2 is an explanatory diagram showing the processing operation of the route search apparatus according to the first embodiment.
As shown in FIG. 1, the route search apparatus includes a camera 1, a human detection unit 2, an attribute estimation unit 3, a GPS 4, a data integration unit 5, a data storage unit 6, an operation input unit 7, a route generation unit 8, and a route display unit 9. It consists of

  As shown in FIGS. 1 and 2, the route search apparatus performs a data collection process and a route search process using the collected data. The data collection process is executed by the camera 1, the human detection unit 2, the attribute estimation unit 3, the GPS 4, the data integration unit 5 and the data storage unit 6, and the route search process is performed by the data storage unit 6 and the low encounter probability priority selection unit 7a. The operation input unit 7, the route generation unit 8, and the route display unit 9 are provided.

Next, the processing operation of the route search device will be described with reference to FIG.
First, the data collection process will be described. The predetermined area A including the pedestrian 12 is photographed using the camera 1 arranged in front of the vehicle 11, and the image 13 is acquired by the photographing. The human detection unit 2 detects the pedestrian 14 in the acquired image 13. The attribute estimation unit 3 estimates the attributes of the detected pedestrians 14 and generates attribute information classified into the elderly, adults, and children. The data integration unit 5 generates integrated data that integrates the generated attribute information with the position information and time information acquired by the GPS 4 mounted on the vehicle 11 and stores the integrated data in the data storage unit 6.

  Next, a route search process using the accumulated data accumulated in the data accumulation unit 6 will be described. When the item “low encounter probability priority” is selected 16 from the items “time priority”, “price priority”, or “low encounter probability priority” displayed on the display screen of the navigation device 15 or the like, the route generation unit 8 Performs route search processing giving priority to a route having a low encounter probability with a pedestrian based on position information and time information acquired by the GPS 4 and accumulated data stored in the data accumulation unit 6. The route display unit 9 displays the travel route set by the route generation unit 8 on the screen.

Next, details of each configuration for executing the data collection processing will be described.
The camera 1 periodically captures the predetermined area A and outputs the captured image 13 to the person detection unit 2 and the attribute estimation unit 3. The human detection unit 2 detects a pedestrian 14 from the input image 13 and outputs image coordinates that are detection results of the pedestrian 14 to the attribute estimation unit 3.
Here, pedestrians are detected by learning appearance information obtained from an image in advance by machine learning and using the learning result. An HOG (Histogram of Oriented Gradient: see Reference 1) feature or the like that uses gradient information as appearance information may be used, and SVM (Support Vector Machine: see Reference 2) or the like may be used for learning.
[Reference 1]
N. Dalal and B. Triggs: `` Histograms of Oriented Gradients for Human Detection '', CVPR, pp. 886-893, 2005.
[Reference 2]
Support Vector Machine: Nello Cristianini (original), John Shawe-Taylor (original), Takeshi Ohkita (translation): "Introduction to Support Vector Machine". Kyoritsu Shuppan, 2005

The attribute estimation unit 3 extracts the image of the pedestrian 14 from the image coordinates input from the human detection unit 2 and the image 13 input from the camera 1. Or classify into three attributes of children. The classification result is output to the data integration unit 5 as attribute information.
Here, the attribute classification is performed by machine learning using appearance information and using the discriminator as in the case of pedestrian detection in the human detection unit 2. The methods disclosed in Reference Document 3 and Reference Document 4 shown below may be applied, and means are not limited.
[Reference 3]
JP 2005-165447 A [Reference 4]
JP 2005-148880 A

  The data integration unit 5 integrates the attribute information input from the attribute estimation unit 3, the position information and time information acquired by the GPS 4, and outputs the integrated data to the data storage unit 6. The integrated data includes, for example, the number of elderly people, the number of adults, the number of children, latitude, longitude, and time. The data storage unit 6 stores the integrated data input from the data integration unit 5.

Next, a configuration for executing the route search process will be described.
The data storage unit 6 stores the integrated data generated by the data collection process. The operation input unit 7 includes input means such as a keyboard, a touch panel, and a mouse, for example, and accepts destination setting, route generation condition specification, and the like. In FIG. 2, the configuration for designating the generation of the low encounter probability priority route is shown as the low encounter probability priority selection unit 7a. The destination information received by the operation input unit 7 and the low encounter probability priority route generation instruction are output to the route generation unit 8.

  When the destination information input from the operation input unit 7 and the low generation probability priority route generation instruction are input, the route generation unit 8 stores data based on the current position information and time information of the vehicle 11 acquired from the GPS 4. A route to the destination is searched by accessing the unit 6, a route having a low encounter probability with a pedestrian is selected from the searched routes, and a travel route as a selection result is output to the route display unit 9. The route display unit 9 displays the travel route input from the route generation unit 8 on a display screen (not shown) such as a monitor of the navigation device 15.

The configuration of the route generation unit 8 will be described in more detail with reference to FIG. The route generation unit 8 includes a route search unit 8a, an encounter probability calculation unit 8b, a weight setting unit 8c, and a route selection unit 8d. The path is generated using the Dijkstra method.
The route search unit 8a accesses the data storage unit 6 based on the destination information input via the operation input unit 7, the current position information and time information of the vehicle 11 acquired from the GPS 4, and searches for a route. .

なお、老人、大人、子供の遭遇確率の算出は、上述した式（３）に限定されるものではなく、時間ごとの確率、あらかじめ設定した距離あたりの確率、およびあらかじめ設定した領域当たりの確率を算出するように構成可能である。 The encounter probability calculation unit 8b calculates the encounter probability of each route searched by the route search unit 8a based on the following equation (1).
Encounter probability = f (old man, adult, child) (1)
In Expression (1), the function f is a function for calculating the encounter probability when an elderly person, an adult, or a child is input, and can be set as shown in Expression (2) below.
f (old man, adult, child) = λ old man + μ adult + ν child (2)
In equation (2), λ, μ, and ν are weights for each attribute. Moreover, the elderly person, the adult, and the child in Expression (2) are probabilities of encounter when passing the road, and are calculated based on the following Expression (3). In the following equation (3), the encounter probability of each attribute is calculated for each route from the searched current position to the destination.

Note that the calculation of the encounter probability of an elderly person, an adult, or a child is not limited to the above-described formula (3), and the probability for each time, the probability for a preset distance, and the probability for a preset area are calculated. It can be configured to calculate.

The weight setting unit 8c calculates the weight w of each route based on the following equation (4) using the encounter probability calculated by the above equations (1) to (3).
w = α cost + (1−α) encounter probability (4)
In Expression (4), α is a weight related to the cost and the encounter probability, and the cost is a weight used in the Dijkstra method.

  The route selection unit 8d refers to the weight w of each route set by the weight setting unit 8c, and selects the route having the smallest value of the weight w as the travel route. As described above, in the Dijkstra method, in order to generate an optimum route, the route is generated by reflecting the integrated data obtained from the data storage unit 6 in the weight w.

Next, the operation of the route search apparatus will be described separately for data collection processing and route search processing.
First, FIG. 3 is a flowchart showing the data collection processing operation of the route search apparatus according to the first embodiment of the present invention.
The camera 1 captures an image and outputs the captured image to the person detection unit 2 and the attribute estimation unit 3 (step ST1). The human detection unit 2 performs a pedestrian detection process in the image based on the captured image input in step ST1 (step ST2). Furthermore, with reference to the detection result of step ST2, it is determined whether or not a pedestrian has been detected (step ST3). When a pedestrian is not detected (step ST3; NO), a process is complete | finished and it transfers to the image acquisition process from a different camera.

  On the other hand, when a pedestrian is detected (step ST3; YES), the detected image coordinates of the pedestrian are acquired and output to the attribute estimation unit 3 (step ST4). The attribute estimation unit 3 extracts the pedestrian image from the photographed image input in step ST1 using the image coordinates input in step ST4, and estimates the pedestrian attribute from the extracted pedestrian image. The information is classified into three types, elderly, adult and child, and the classified information is output to the data integration unit 5 as attribute information (step ST5).

  The data integration unit 5 acquires position information and time information from the GPS 4 (step ST6), generates integrated data that integrates the acquired position information and time information, and the attribute information input in step ST5, and stores the data. It outputs to the part 6 (step ST7). The data storage unit 6 stores the integrated data input in step ST7 (step ST8) and ends the process.

Next, route search processing of the route search device will be described. FIG. 4 is a flowchart showing the data collection processing operation of the route search apparatus according to the first embodiment of the present invention.
When the operation input unit 7 receives an input operation for instructing setting of a destination (step ST11), the operation input unit 7 further determines whether or not the low encounter probability priority selection unit 7a has received designation of a route search with low encounter probability priority (step ST11) ST12). If the selection instruction for the low encounter probability priority route search is not received (step ST12; NO), the process is terminated. On the other hand, when an instruction for selecting a route search with a low encounter probability is accepted (step ST12; YES), the route search instruction with a low encounter probability priority is output to the route search unit 8 (step ST13).

  The route search unit 8 acquires position information and time information from the GPS 4 according to the low encounter probability priority route search instruction input in step ST13 (step ST14). The acquired position information and time information and the data storage unit 6 Are searched for a route having a low encounter probability with a pedestrian, and the generated route is output to the route display unit 9 (step ST15). The route display unit 9 displays the route input in step ST15 on a display screen such as a monitor (step ST16), and ends the process.

Details of the route generation processing in step ST15 will be described with reference to FIGS.
FIG. 5 is a flowchart showing the operation of the route search unit of the route search device according to Embodiment 1 of the present invention. FIG. 6 is a diagram showing the map information as a graph. FIG. 7 is an explanatory diagram showing an example of setting an optimum route based on the map information shown in FIG.
The path is generated using the Dijkstra method as described above. 6, the current vehicle position is indicated by C (hereinafter referred to as current position C), the destination is indicated by G (hereinafter referred to as destination G), and each intersection is indicated by a number from 1 to 4. The roads are indicated by straight lines connecting the intersections. The weight w attached to the straight line indicating the road is a numerical value determined by the distance or the road width of the road, and is calculated based on the above-described equation (4).

A route from the current position C of the vehicle to the reachable intersection is set, and the weight w of the set route is calculated (step ST21). In the example shown in FIG. 7 (a), as the processing in step ST21, sets the route C1 (configuration path) and the path C2, and calculates a weight _{w c2} weights _{w c1} and path C2 path C1. Next, a further reachable intersection is searched from the intersection reached in step ST21, and the weight w of the route to the searched intersection is added (step ST22). In the example illustrated in FIG. 7B, as the process of step ST22, routes C12, C13, and C14 are set, the weight w _c1 + w _{12 of} the route C12, the weight w _c1 + w _{13 of the} route C13, and the weight w _{c2 of the} route C24. + W ₂₄ is calculated.

  Next, the routes are rearranged in order of decreasing weight w added in step ST22 (step ST23). In the example shown in FIG. 7C, the process in step ST23 is rearranged in the order of routes C13, C24, and C12. Next, it is determined whether the destination G is included in the route rearranged in step ST23 (step ST24). When the destination G is not included in the route (step ST24; NO), the processing returns to step ST22 and the above-described processing is repeated.

  On the other hand, when the destination G is included in the route (step ST24; YES), the route having the smallest weight w among the routes rearranged in step ST23 is set as the optimum route (step ST25), and the route search process is performed. finish. In the example shown in FIG. 7C, since it is determined in step ST24 that the destination C is not included in the routes C13, C24, and C12, the process returns to step ST22 to search for a further reachable intersection. Then, routes C134, C13G, C24G, and C124 are set, and the weight w of each route is calculated (see FIG. 7D). Next, in step ST23, the routes C13G, C24G, C134, and C124 are rearranged in order, and in step ST24, it is determined that the destination C is included in the route C13G and the route C24G. The small route C13G is set as the optimum route (see FIG. 7E), and the process is terminated.

  As described above, according to the first embodiment, the person detection unit 2 that detects a pedestrian from a captured image, the attribute estimation unit 3 that extracts a pedestrian image and classifies it into preset attributes, Since it comprises the data integration part 5 which produces | generates the integrated data which integrated attribute information, position information, and time information, and the data storage part 6 which accumulate | stores integrated data, it is the structure of the walking vehicle information on the road in driving | running | working Accumulation is possible.

  Further, according to the first embodiment, when a low encounter probability priority route generation instruction is input, a route with a low encounter probability with a pedestrian is referred to with reference to the integrated data stored in the data storage unit 6. Since the route generation unit 8 to be searched is provided, a route with a low encounter probability with a pedestrian can be generated based on the accumulated pedestrian information. Thereby, the stress which a driver | operator receives by encounter with a pedestrian can be reduced. Moreover, the occurrence of a contact accident with a pedestrian can be prevented.

  In the first embodiment described above, a configuration is shown in which pedestrians are classified into the three attributes of an elderly person, an adult, and a child. However, the attributes to be classified are not limited to these three and can be changed as appropriate. .

  In the first embodiment described above, a configuration is shown in which a pedestrian is detected from a captured image and data is collected. However, the detection of a pedestrian may be manually input.

  In Embodiment 1 described above, the method using the HOG feature is described as a method for detecting a pedestrian, but the method is not limited to this method. Moreover, although SVM was mentioned as a learning method, it is not limited to the said method.

In the first embodiment described above, the function f is calculated by assigning weights w to the probabilities of the elderly, adults, and children. However, as shown in the following equation (5), the elderly, adults, and children are calculated. All probabilities may be combined and used as a pedestrian encounter probability.

Embodiment 2. FIG.
In this Embodiment 2, in addition to the structure of Embodiment 1 mentioned above, the route search apparatus provided with the structure which measures the distance to the pedestrian and object ahead of a vehicle is shown.
FIG. 8 is a block diagram showing the configuration of the route search apparatus according to Embodiment 2 of the present invention, and FIG. 9 is an explanatory diagram showing the processing operation of the route search apparatus according to Embodiment 2 of the present invention. A distance sensor 10 is additionally provided in the route search apparatus shown in the first embodiment. In the following description, the same or corresponding parts as the components of the route search apparatus according to the first embodiment are denoted by the same reference numerals as those used in the first embodiment, and description thereof is omitted or simplified.

  As shown in FIG. 2, the distance sensor 10 is disposed in front of the vehicle 11 and measures the distance to a pedestrian or object located in front of the vehicle 11. The human detection unit 2 detects the pedestrian 14 in the image 13 using the image acquired by the camera 1 and the distance data acquired by the distance sensor 10. Since the other configuration of the route search apparatus is the same as that of the first embodiment, description thereof is omitted.

Next, the operation of the data collection process of the route search device according to the second embodiment will be described.
FIG. 10 is a flowchart showing the data collection processing operation of the route search apparatus according to the second embodiment of the present invention. In the following, the same steps as those of the data collection process of the route search apparatus according to the first embodiment are denoted by the same reference numerals as those used in FIG. 3, and the description thereof is omitted or simplified.
The camera 1 captures an image and outputs the captured image to the person detection unit 2 and the attribute estimation unit 3 (step ST1). At the same time as the process of step ST1, the distance sensor 10 measures the distance to a pedestrian or object located in front of the vehicle, acquires distance data, and outputs the distance data to the human detection unit 2 and the attribute estimation unit 3 (step ST31). ).

  The human detection unit 2 performs a pedestrian detection process in the image based on the captured image and the distance data input in step ST1 and step ST31 (step ST2 ′). Further, it is determined whether or not a pedestrian has been detected with reference to the detection result of step ST2 ′ (step ST3). When a pedestrian is not detected (step ST3; NO), a process is complete | finished and it transfers to the image acquisition process from a different camera.

  On the other hand, when a pedestrian is detected (step ST3; YES), the detected image coordinates of the pedestrian are acquired and output to the attribute estimation unit 3 (step ST4). The attribute estimation unit 3 uses the image coordinates input in step ST4 and the distance information input in step ST31 to extract a pedestrian image from the captured image input in step ST1, and extracts the extracted pedestrian's image. The attributes of the pedestrian are estimated from the image and classified into three types, elderly, adult and child, and the classified information is output to the data integration unit 5 as attribute information (step ST5). Since the subsequent processing is the same as that of the first embodiment, description thereof is omitted.

  In the pedestrian detection process in the human detection unit 2 according to the second embodiment, the appearance information obtained from the image and the distance to the pedestrian obtained from the distance data are learned in advance by machine learning, and the learning result is used. Do. By using the distance data obtained by the distance sensor 10, not only the pedestrian texture information but also features such as the distance to the pedestrian and the size of the pedestrian can be used. It can be performed. As in the first embodiment, an HOG feature that uses gradient information as appearance information may be used, and SVM may be used for learning.

  Further, the classification by the attribute estimation unit 3 is performed by machine learning using appearance information and distance information, and by the discriminator, similarly to detection of a pedestrian. Similar to pedestrian detection, this classification can use features such as the size of a pedestrian or an object by using distance information obtained by the distance sensor 10, and therefore can perform identification with high accuracy. It becomes possible. Further, as the classification method, the methods disclosed in Reference Document 3 and Reference Document 4 described in Embodiment 1 may be applied, and means are not limited.

  As described above, according to the second embodiment, the distance sensor 10 that measures the distance to the pedestrian or object ahead of the host vehicle is provided, and the human detection unit has the captured image of the camera 1 and the distance between the distance sensor 10. Since the pedestrian is detected using the data, it is possible to identify the pedestrian with high accuracy. Also, a route with a low encounter probability with a pedestrian can be generated based on highly accurate pedestrian information. Thereby, the stress which a driver | operator receives by encounter with a pedestrian can be reduced. Furthermore, the occurrence of a contact accident with a pedestrian can be prevented.

  In the present invention, within the scope of the invention, any combination of the embodiments, or any modification of any component in each embodiment, or omission of any component in each embodiment is possible. .

  1 camera, 2 person detection unit, 3 attribute estimation unit, 4 GPS, 5 data integration unit, 6 data storage unit, 7 operation input unit, 7a low encounter probability priority selection unit, 8 route generation unit, 8a route search unit, 8b Encounter probability calculation unit, 8c weight setting unit, 8d route selection unit, 9 route display unit, 10 distance sensor, 11 vehicle, 12, 14 pedestrian, 13 images, 15 navigation device, 16 selection.

Claims (3)

A human detection unit that detects a pedestrian from a captured image of a predetermined area from the host vehicle;
A data integration unit that generates integrated data integrating the pedestrian information detected by the person detection unit and the vehicle position information;
A data storage unit for storing integrated data generated by the data integration unit;
An operation input unit that accepts destination setting and route generation condition specification;
When a route generation condition with a low encounter probability with a pedestrian is specified via the operation input unit, a route for searching for a route from the current position of the host vehicle to the set destination based on the specification An encounter probability calculating unit for calculating an encounter probability of a constituent route constituting the route searched by the route search unit, using the search unit, the integrated data stored in the data storage unit, and the vehicle position information; , Based on the encounter probability calculated by the encounter probability calculation unit, a weight setting unit that sets the weight of the configured route, and the weight of the configured route set by the weight setting unit, and among the searched routes A route generation unit including a route selection unit that selects a route having the smallest weight as a travel route ;
An attribute estimation unit that estimates the attribute of the pedestrian from the captured image and the pedestrian information detected by the person detection unit;
The data integration unit generates the integrated data by integrating the attribute information of the pedestrian estimated by the attribute estimation unit and the vehicle position information.
The encounter probability calculation unit calculates an attribute-specific encounter probability for each route for each attribute information of the pedestrian, and applies a preset weight for each attribute of the pedestrian to the calculated attribute-specific encounter probability And calculating the encounter probability . A distance sensor for measuring the distance from the vehicle to the pedestrian,
The route search device according to claim 1, wherein the human detection unit detects a pedestrian from the captured image using a measurement result of the distance sensor. The vehicle position information, the route searching apparatus according to claim 1 or claim 2 wherein, characterized in that it is constituted by the position information and time information of the own vehicle.

Images