JP2007279970A - Object detection apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an object detection apparatus for detecting objects with high accuracy regardless of the number of objects to be detected. <P>SOLUTION: The object detection apparatus 1 for detecting objects from images picked up by an image pickup means 3 includes a detection criteria setting means 5a for setting detection criteria based on the number of objects to be detected, and an object detecting means 5b for detecting the objects to be detected from the images according to the detection criteria set by the detection criteria setting means 5a. When the number of objects to be detected is large, the detection criteria setting means 5a sets less strict detection criteria than when the number is small. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an object detection apparatus that detects an object from a captured image.

In order to assist a driver of a vehicle, an apparatus for providing information such as a pedestrian is known. In the pedestrian detection apparatus shown in Patent Document 1, a pedestrian is detected from an image captured by an imaging unit, and the pedestrian detected in the captured image is emphasized and displayed by a display device. In particular, in the case of a night view system, by detecting a pedestrian from an infrared image captured by an infrared camera and highlighting the detected pedestrian on a display device, the pedestrian is invisible to the naked eye of the driver when driving at night. Information is also provided. As a method for detecting a pedestrian, for example, a region having a predetermined size is cut out from a captured image, and whether or not the user is a pedestrian is determined by pattern matching between the image of the region and the template of the pedestrian. In the determination of whether or not the person is a pedestrian, the determination is performed by comparing the matching degree (similarity) by pattern matching with a parameter (threshold value).
JP 2001-28050 A

  In the case of the conventional pedestrian detection method, the parameters used for detection are not optimally set according to the situation of the pedestrian, so that erroneous detection or detection omission may occur. For example, when a parameter with a high detection rate is set, it may be erroneously detected that a pedestrian exists even in a situation where no pedestrian exists. In this case, since the pedestrian is not present, it is highlighted on the display device. Therefore, the driver gazes at the screen and takes an avoidance action while anxious about what is reflected. On the other hand, if a detection rate parameter is set such that no false detection occurs, detection omission may occur in a crowded situation where there are many pedestrians. In this case, although the pedestrian is present, the pedestrian is not highlighted on the display device, so the driver feels distrust of the system.

  Therefore, an object of the present invention is to provide an object detection apparatus that detects an object with high accuracy regardless of the number of detection target objects.

  An object detection device according to the present invention is an object detection device that detects an object from a captured image captured by an imaging unit, and includes a detection criterion setting unit that sets a detection criterion based on the number of detection target objects, and a detection criterion setting Object detection means for detecting the detection target object from the captured image based on the detection reference set by the means, and the detection reference setting means is a relaxed detection reference compared to the case where the number of detection target objects is large compared to the case where the number is small Is set.

  In this object detection device, the detection reference setting means sets a detection reference used for object detection based on the number of detection target objects, and the object detection means detects the object from the captured image using the set detection reference. In particular, the detection criterion setting means sets a detection criterion that is more relaxed than usual when the number of detection target objects is large compared to when the number is small. In other words, a predetermined detection standard is set during normal operation to prevent false detection in places where the number of detection target objects is small (especially when there are no detection target objects at all) and detect an object at a normal detection rate. To do. The predetermined detection standard is, for example, a balanced value considering erroneous detection and detection omission. On the other hand, when the number of detection target objects increases, a detection criterion that is relaxed compared to normal times is set, the detection rate is increased, and detection omissions at places where the number of detection target objects is large are suppressed. As described above, this object detection apparatus can detect an object with high accuracy regardless of the number of detection target objects. By the way, when the number of detection target objects is large, the area ratio of the detection target object to the captured image becomes large. Therefore, the area ratio that may cause a false detection in the captured image (that is, the area ratio where the detection target object does not exist) ) Decreases. Therefore, even if the false detection rate increases by relaxing the parameters, the false detection rate as the object detection device is suppressed.

  The number of detection target objects used when setting the detection reference may be, for example, the number of detection target objects detected at the same location in the past, or a detection target detected immediately before that location. It may be the number of objects. The number of detection target objects detected in the same place in the past may use a past detection result by the object detection device itself, or may be a detection result by other detection means. As another detection means, for example, a detection result by a person or a detection result by a monitoring device installed on the road is used, and the detection result is held in the object detection device.

  In the object detection apparatus of the present invention, the detection reference setting means may set a more relaxed detection reference as the number of detection target objects is larger.

  In this object detection apparatus, the detection reference setting means sets a more relaxed detection reference as the number of detection target objects increases. That is, a plurality of stages to be relaxed are prepared as detection standards, and a detection standard having a greater relaxation degree is set as the number of detection target objects increases. Accordingly, the object detection apparatus can set a more appropriate detection reference according to the number of detection target objects, and can further suppress detection omission when the number of detection target objects increases.

  The object detection device of the present invention further includes detection information holding means for holding at least the number information of the detection target objects and the location information where the detection target object is detected, and the detection reference setting means includes the detection information holding means. The detection criterion may be set with reference to the information held in step (b).

  In this object detection apparatus, the number information of the detection target objects detected in the past and the location information where the detection target object is detected are held in association with each other in the detection information holding means. In the object detection device, the detection reference setting means refers to the held information, and sets the detection reference based on the number information of the past detection target objects corresponding to the place where the current detection is performed. As described above, the object detection apparatus can estimate the number of detection target objects at the place where the current detection is performed by using the past detection result, and set an appropriate detection reference.

  In addition to the location information that most affects the increase / decrease in the number of detection target objects, the information held in association with the number information of detection target objects includes the day of the week, holidays / weekdays, seasons, time zones, weather, temperature, Various information that affects the number of detection target objects such as scenes (city areas, mountains, countryside, highways, etc.) may be added.

  In the object detection apparatus of the present invention, the object detection means determines that the object is a detection target object when the evaluation value at the time of object detection exceeds the detection reference, and the detection reference setting means has a large number of detection target objects. In some cases, a smaller value is set as a detection reference than when the number is small.

  In this object detection device, an evaluation value (for example, matching degree in the case of pattern matching) is obtained by the object detection means, and when the obtained evaluation value exceeds a detection criterion, it is determined as a detection target object. In the object detection device, the detection reference setting means sets a small value as the detection reference when the detection reference is relaxed. Thereby, in a place where the number of detection target objects is large, the detection target object is detected even with a smaller evaluation value than in a place where the number of detection target objects is small, and the detection rate is increased.

  In the object detection device of the present invention, the object detection means determines that the object is a detection target object when the evaluation value when performing object detection is below the detection reference, and the detection reference setting means has a large number of detection target objects. In some cases, a larger value is set as the detection reference than when the number is small.

  In this object detection apparatus, an evaluation value is obtained by the object detection means, and when the obtained evaluation value falls below a detection criterion, it is determined as a detection target object. In the object detection device, when the detection standard is relaxed, the detection standard setting means sets a large value as the detection standard. Thereby, in a place where the number of detection target objects is large, the detection target object is detected even with a larger evaluation value than in a place where the number of detection target objects is small, and the detection rate is increased.

  In the present invention, by setting the detection reference in consideration of the number of detection target objects, erroneous detection and detection omission can be suppressed, and an object can be detected with high accuracy.

  Hereinafter, an embodiment of an object detection device according to the present invention will be described with reference to the drawings.

  In the present embodiment, the object detection device according to the present invention is applied to a night view system mounted on a vehicle. The night view system according to the present embodiment detects a pedestrian from an infrared image (invisible image) obtained by an infrared camera and provides pedestrian information existing in front of the vehicle in order to assist the driver during night driving. . In this embodiment, there are two forms depending on the difference in the method of acquiring the number of pedestrians, and the first embodiment uses a detection result when the past travels. Is a form that uses the detection result in the previous frame that is currently running.

  With reference to FIG. 1, the night view system 1 which concerns on 1st Embodiment is demonstrated. FIG. 1 is a configuration diagram of a night view system according to the first embodiment.

  The night view system 1 detects a pedestrian from an infrared image and displays information indicating the detected pedestrian. In the night view system 1, in order to suppress false detection in a place with few pedestrians and detection omission in a place with many pedestrians, parameters (detection criteria) used in pedestrian detection are Variable depending on the crowd situation (number of pedestrians). In particular, in the night view system 1, the crowd information is accumulated by using the detection result of the pedestrian when traveling in the past, and the crowd situation of the place where the current detection is performed is estimated from the accumulated crowd information. The night view system 1 includes an infrared projector 2, an infrared camera 3, a display device 4, and an ECU [Electronic Control Unit] 5, and uses a navigation system 6.

  The infrared projector 2 is disposed at the front end of the vehicle and attached toward the front of the vehicle. The infrared projector 2 irradiates the front of the vehicle with infrared rays (for example, near infrared rays). The infrared projector 2 is turned on when the night view system 1 is started and turned off when the night view system 1 is stopped. In addition, when infrared rays are near infrared rays, since the near infrared rays are contained in the light irradiated from a headlight, it is good also as a structure which does not have the infrared projector 2. FIG.

  The infrared camera 3 is disposed on the front side of the vehicle (for example, the back side of the rearview mirror) and attached toward the front of the vehicle. The infrared camera 3 takes in infrared rays (such as infrared reflected light from the infrared projector 2), and generates an infrared video image by using light and shade according to the intensity of the infrared rays. This infrared image is composed of an infrared image of a frame every predetermined time (for example, 1/30 second). The infrared camera 3 transmits infrared image information of each frame to the ECU 5 as an image signal at regular time intervals.

  The display device 4 is a display unit that provides the driver with infrared images and detected pedestrian information, and is, for example, a liquid crystal display or a head-up display used in the navigation system 6. The display device 4 receives a display signal from the ECU 5 and displays an image indicated by the display signal. FIG. 2 shows an example of a display image when displayed on the display device 4 (liquid crystal display), and the detected pedestrian is surrounded by a rectangular line and highlighted. FIG. 2A shows a case where there is no detection omission, and all pedestrians in the image are surrounded by rectangular lines. On the other hand, FIG. 2B shows a case where there is a detection omission, and the upper right pedestrian in the image is not surrounded by a rectangular line.

  The navigation system 6 is a system that performs detection of the current position and traveling direction of the host vehicle, route guidance to a destination, and the like. In particular, in the navigation system 6, every time the current position is detected by the current position detector 6a, the current position information is transmitted to the ECU 5 as a current position signal. When the navigation system 6 receives the map information request signal from the ECU 5, the navigation system 6 reads map information and environment information around the current position from the map / environment information database 6b, and transmits the read information to the ECU 5 as a map information signal. When the navigation system 6 receives a garbage information request signal from the ECU 5, the navigation system 6 reads the garbage information around the current position from the garbage information database 6c, and transmits the read garbage information to the ECU 5 as a garbage information signal. Further, when the navigation system 6 receives a garbage information accumulation signal consisting of newly generated garbage information from the ECU 5, the navigation system 6 stores the garbage information in the garbage information database 6c. In the first embodiment, the garbage information database 6c corresponds to detection information holding means described in the claims.

  The environmental information is various information regarding each place shown on the map, and is scene information of places such as a factory district, a commercial district, a residential district, an urban area, a mountain, a countryside, and a highway. When the navigation system 6 does not hold environment information, the ECU 5 acquires only map information. The garbage information database 6c may be configured in the ECU 5 when the navigation system 6 does not use the garbage information. Further, when the navigation system 6 transmits the current position information without transmitting the map information request signal or the crowd information request signal from the ECU 5, the map information, the environment information, and the crowd information around the current position are automatically transmitted. You may make it do.

  The ECU 5 includes a CPU [Central Processing Unit], a ROM [Read Only Memory], a RAM [Random Access Memory], and the like. By executing software held in the ROM by the CPU, the parameter adjustment unit 5a and the pedestrian detection unit 5b. The detection result analysis unit 5c is configured. The ECU 5 receives an image signal from the infrared camera 3 at regular intervals and detects a pedestrian from the infrared video (infrared image). Then, the ECU 5 displays the infrared image and the detection result of the pedestrian on the display device 4. In particular, the ECU 5 generates the garbage information from each information from the navigation system 6 and the detection result of the pedestrian, and causes the navigation system 6 to accumulate the garbage information. Then, the ECU 5 extracts the crowd information at the current position from the accumulated crowd information, and sets parameters used for pedestrian detection according to the crowd information. In the first embodiment, the parameter adjustment unit 5a corresponds to the object reference setting unit described in the claims, and the pedestrian detection unit 5b corresponds to the object detection unit described in the claims.

  When the parameter adjustment unit 5a acquires the current position information from the navigation system 6, the parameter adjustment unit 5a transmits a person information request signal for requesting the person information at the current position, and acquires the person information at the current position from the navigation system 6. Then, the parameter adjustment unit 5a determines whether or not the number of pedestrians detected in the past at the current position is equal to or greater than the number of crowd determinations. If the number is less than the number of crowd determinations, a normal parameter is set and the number of crowd determinations is equal to or greater. In the case of, set the parameters for crowd.

  The number of crowd determinations is the minimum number of pedestrians in which a detection failure may occur, and is set in advance by an experiment or the like. The normal parameter is a parameter that is used for pedestrian detection during normal times (when the number of pedestrians is small), and is a parameter that provides a balanced detection rate in consideration of suppression of erroneous detection and detection omission. The crowd parameter is a parameter used for pedestrian detection when the number of pedestrians is large (as a result, there is a possibility that detection failure may occur), and is a parameter for increasing the detection rate. Accordingly, the crowd parameter is a parameter that makes it easy for pedestrians to be detected, and is a parameter that is relaxed from the normal parameter. Each parameter is preset by an experiment or the like.

  The parameter for the crowd may be one level, or may be a parameter having a plurality of levels in which the degree of relaxation increases as the number of pedestrians increases. For example, as the parameters for crowds, three levels are prepared for a small number of people, a medium number of people, and a large number of people according to an increase in the number of pedestrians, and are gradually reduced.

  In the pedestrian detection part 5b, a pedestrian is detected from the infrared image from the infrared camera 3 using the normal parameter or the crowd parameter set by the parameter adjustment part 5a. The method for detecting a pedestrian is not particularly limited, and various methods can be applied. Accordingly, the parameters also vary depending on the detection method, and the method of relaxing the parameters also varies depending on the detection method.

  As a detection method, for example, there is a method of preparing a pedestrian template and pattern matching using the template. Specifically, an area of a predetermined size (pedestrian candidate area) is sequentially cut out from the infrared image, a matching degree (evaluation value) between the template image and each cut out pedestrian candidate area image is obtained, and the matching degree If is greater than or equal to the threshold, the pedestrian candidate is determined to be a pedestrian. At this time, a template for the entire pedestrian may be prepared and the pedestrian may be detected by one-step pattern matching, or a plurality of partial templates such as the pedestrian's head, legs, arms, and torso may be prepared. A pedestrian may be detected by pattern matching in a plurality of stages. In the case of multiple stages, pattern matching is performed for each pedestrian candidate in each stage, and the pedestrian candidate moves to pattern matching in the next stage only when the matching degree of each pedestrian candidate is greater than or equal to a threshold value. A pedestrian candidate whose matching degree is equal to or greater than a threshold value at all stages is determined to be a pedestrian. In the case of such pattern matching, the threshold value and the number of steps for determining the matching degree are parameters. In order to relax the parameters, the smaller the value in the case of the threshold, the more relaxed, and the smaller the number in the number of stages, the more relaxed.

  When the detection is completed, the pedestrian detection unit 5b generates a display image in which the detected pedestrian is emphasized in the infrared image (monochrome grayscale image), and transmits the display image information to the display device 4 as a display signal. As a display method for emphasizing the pedestrian, for example, the pedestrian in the infrared image is displayed surrounded by a rectangular line (see FIG. 2).

  The detection result analysis unit 5 c transmits a map information request signal for requesting map information at the current position, and acquires map information and environment information from the navigation system 6. Then, the detection result analysis unit 5c generates crowd information by associating the number of pedestrians detected by the pedestrian detection unit 5b with map information and environment information of the current position and the current position. Further, the detection result analysis unit 5c transmits a garbage information accumulation signal including the garbage information to the navigation system 6 and accumulates it in the garbage information database 6c.

  The increase or decrease in the number of pedestrians (the situation of crowds) is a significant factor in location, but the number of pedestrians varies depending on various factors such as time of day and day of the week. Therefore, as information associated with the number of pedestrians as the crowd information, not only the location information (current position and the like) but also various information that causes the crowd situation to change may be associated. Examples of information that causes changes in crowds include day of the week, holidays / weekdays, seasons, time zones, weather, and temperature. As described above, when information other than the location information is added to the crowd information, the parameter adjustment unit 5a sets the parameters in consideration of the information. For example, in the past detection, if there is 3% or more of pedestrians on a weekday morning on a sunny day in a certain place, if 100% of the pedestrians are present, and if more than 90% of the data is accumulated, When three or more pedestrians are detected on a weekday morning on a sunny day at the place, a parameter for crowding is set. In this way, by adding a factor that changes the crowd situation as the crowd information, it is possible to estimate the crowd situation with higher accuracy and more detail.

  In addition, when traveling again in the same place that has traveled in the past, new garbage information is accumulated in addition to the already accumulated garbage information about the place. Thus, by accumulating the crowd information for the same place, it becomes possible to estimate the crowd situation for the place with higher accuracy.

  Note that the ECU 5 performs pedestrian detection using the normal parameters in the background when pedestrian detection is performed using the crowd parameters, and the detection results of the two parameters are compared and verified. Good. In this case, when the ECU 5 determines that it is not necessary to use the parameter for the crowd from the verification result, the ECU 5 switches the parameter for the crowd to the normal parameter. Such a process is effective when the parameters for the garbage are used at an early stage from the stage where the garbage information is not yet accumulated, or when the reliability of the accumulated garbage information is low.

  The operation of the night view system 1 will be described with reference to FIG. In particular, the processing in the ECU 5 will be described along the flowchart of FIG. FIG. 3 is a flowchart showing a process flow in the ECU of the night view system according to the first embodiment.

  The infrared camera 3 takes infrared rays to form infrared images, and transmits image signals of infrared image information of each frame to the ECU 5 at regular intervals. The navigation system 6 detects the current position and transmits the current position signal to the ECU 5.

  The ECU 5 receives an image signal from the infrared camera 3 and acquires an infrared image at regular intervals (S1). Further, the ECU 5 receives the current position signal from the navigation system 6 and acquires the current position (S2). Further, the ECU 5 receives the garbage information signal from the navigation system 6 and acquires the garbage information around the current position (S3).

  Then, the ECU 5 sets parameters used for pedestrian detection based on the current position and the crowd information (the number of pedestrians detected in the past) at the current position (S4). Here, when the current position can be estimated from the past detection result, the parameter for the crowd is set, and when it is estimated that the current position is not the crowd, the normal parameter is set.

  Next, the ECU 5 detects a pedestrian from the infrared image by pattern matching or the like using the set parameters (S5). Here, the detection rate of pedestrians increases when the parameter for crowding is set, and the erroneous detection is suppressed by the balance type detection rate when the parameter for normal use is set.

  Furthermore, the ECU 5 generates the garbage information in which the detected number information of the pedestrians and the location information such as the current position are associated with each other, and transmits a garbage information accumulation signal to the navigation system 6 (S6). When this garbage information storage signal is received, the navigation system 6 stores the garbage information in the garbage information database 6c.

  Moreover, in ECU5, the display image which shows the detection result of a pedestrian in an infrared image is produced | generated, and the display signal is transmitted to the display apparatus 4 (S7). When this display signal is received, the display device 4 displays an image in which the detected pedestrian is emphasized in the infrared image.

  According to the night view system 1, by setting parameters used for pedestrian detection according to the number of pedestrians at the current position (especially by relaxing the parameters in a place where the number of pedestrians is large). In addition, it is possible to suppress detection omissions in a place where there are many pedestrians and to suppress erroneous detection in places where there are few pedestrians. As a result, the detection accuracy of pedestrians as a whole system can be improved, and more accurate pedestrian information can be provided to the driver. Therefore, it is possible to eliminate the driver's sense of anxiety due to omission of detection or erroneous detection, and it is possible to improve safety by accurate pedestrian information.

  Furthermore, according to the night view system 1, the detection result of the pedestrian and the detected location information are accumulated as the crowd information, so that the crowd situation (the number of pedestrians) at the time of traveling the place again can be obtained. It is possible to estimate with high accuracy, and it is possible to set parameters according to the crowd situation.

  In addition, when the parameter for garbage is used, the detection rate increases, but the false detection rate also increases. However, in the case of crowds, since the area ratio of pedestrians in the image increases, the area ratio in which erroneous detection may occur in the image (that is, the area ratio in which there is no pedestrian) decreases. Therefore, if ΔERROR (the amount of change in the false detection rate as a system) = Δerr (the amount of change in the false detection rate that varies depending on the parameter) × ΔS (the amount of change in the area where erroneous detection may occur) When Δ is used, Δerr increases, but ΔS decreases. Therefore, even when a parameter for garbage is used, an increase in ΔERROR can be suppressed.

  Incidentally, even if a false detection occurs in a crowd, the annoyance for a driver is significantly different from a false detection that occurs in a place where there are no pedestrians. For this reason, it is more important for the driver that no detection omission occurs in the crowd, and eliminating the omission of detection increases the value of the system.

  Next, the night view system 11 according to the second embodiment will be described with reference to FIG. FIG. 4 is a configuration diagram of the night view system according to the second embodiment. In addition, in the night view system 11, the same code | symbol is attached | subjected about the structure similar to the night view system 1 which concerns on 1st Embodiment, and the description is abbreviate | omitted.

  The night view system 11 is the same system as the night view system 1, but the method for obtaining the number of pedestrians when setting the parameters (the method for detecting the crowd situation) is different. In the night view system 11, the number of pedestrians detected from the infrared image of the previous frame is held, and the crowd situation of the place where the current detection is performed is estimated from the number of pedestrians. The night view system 11 includes an infrared projector 2, an infrared camera 3, a display device 4, and an ECU 15.

  The ECU 15 includes a CPU, a ROM, a RAM, and the like, and a parameter adjustment unit 15a, a pedestrian detection unit 15b, and a detection result holding unit 15c are configured by executing software held in the ROM by the CPU. In the ECU 15, a pedestrian is detected from the infrared image, and the infrared image and the detection result of the pedestrian are displayed on the display device 4, similarly to the ECU 5 according to the first embodiment. In particular, the ECU 15 holds the number of pedestrians detected in each frame each time pedestrian detection is performed on the infrared image of each frame. Then, the ECU 15 sets parameters used for pedestrian detection in the current frame according to the number of pedestrians held in the previous frame. In the second embodiment, the parameter adjustment unit 15a corresponds to the object reference setting unit described in the claims, and the pedestrian detection unit 15b corresponds to the object detection unit described in the claims.

  The parameter adjustment unit 15a determines whether or not the number of pedestrians detected in the previous frame held by the detection result holding unit 15c is equal to or greater than the number of crowd determinations. If the number is greater than the number of people, the parameter for people is set. The interval between the frames of the infrared image is a very short interval of about 1/30 seconds, and the distance that the vehicle can travel at the interval between the frames is very short. Therefore, it can be estimated that the number of pedestrians (the crowd situation) hardly changes between the previous frame and the current frame. In view of this, the number of pedestrians detected in the previous frame is used to estimate the crowd situation in the current frame.

  The pedestrian detection unit 15b performs the same processing as the pedestrian detection unit 5b according to the first embodiment, and each time an infrared image of the current frame is input, the normal parameters set by the parameter adjustment unit 15a or A pedestrian is detected from the infrared image using the parameter for crowd.

  The detection result holding unit 15c holds the number of pedestrians detected in the current frame every time the pedestrian detection unit 15b performs pedestrian detection. The number of pedestrians held is used when setting parameters in the next frame.

  The operation of the night view system 11 will be described with reference to FIG. In particular, the processing in the ECU 15 will be described along the flowchart of FIG. FIG. 5 is a flowchart showing a process flow in the ECU of the night view system according to the second embodiment.

  As in the first embodiment, the infrared camera 3 transmits an image signal of infrared image information of each frame to the ECU 15 at regular intervals, and the navigation system 6 transmits a current position signal to the ECU 15.

  The ECU 15 receives an image signal from the infrared camera 3 at regular intervals, and acquires an infrared image of the current frame (S11).

  The ECU 15 sets parameters used for pedestrian detection based on the number of pedestrians detected in the previous frame held (S12). Here, when the current position can be estimated as crowded from the number of pedestrians detected immediately before, a crowd parameter is set, and when it can be estimated that it is not crowded, a normal parameter is set.

  Next, the ECU 15 detects a pedestrian from the infrared image by pattern matching or the like using the set parameters (S13). Then, the ECU 15 holds the number of pedestrians detected in the current frame (S14). Further, as in the first embodiment, the ECU 15 transmits a display signal to the display device 4 (S15), and the display device 4 displays an image that emphasizes the detected pedestrian in the infrared image.

  According to this night view system 11, as with the night view system 1, it is possible to suppress detection omissions in places where there are many pedestrians and to suppress false detections in places where there are few pedestrians. Pedestrian detection accuracy can be improved. Furthermore, according to the night view system 11, it is possible to estimate the crowd situation of the current frame with high accuracy by a very simple configuration of holding the detection result of the pedestrian, and setting of parameters according to the crowd situation is possible. It becomes possible.

  As mentioned above, although embodiment which concerns on this invention was described, this invention is implemented in various forms, without being limited to the said embodiment.

  For example, in this embodiment, the present invention is applied to pedestrian detection based on an infrared image captured by an infrared camera, but is also applied to pedestrian detection based on an invisible image captured by an invisible camera other than infrared or a visible image captured by a visible camera. The detection target object can also be applied to various objects other than pedestrians. In this embodiment, the present invention is applied to a night view system mounted on a vehicle. However, the present invention can be applied to an in-vehicle device that detects an object other than the night view system, or is not limited to the in-vehicle device. It is applicable to various uses.

  Further, in this embodiment, when the evaluation value for the detection candidate exceeds the parameter as the determination method for the pedestrian, the detection candidate is determined to be a pedestrian. However, when the evaluation value for the detection candidate falls below the parameter Alternatively, the detection candidate may be determined as a pedestrian. In this case, the parameter is relaxed by increasing the value of the parameter.

  Moreover, in 1st Embodiment, although it was set as the structure which accumulate | stores the garbage information based on the detection result of the pedestrian performed by the night view system itself during vehicle driving | running | working in a garbage information database, the detection result of the pedestrian performed by the person Or it is good also as a structure which produces | generates garbage information from the detection result of the pedestrian performed with other apparatuses, such as a periphery monitoring apparatus on a road, and hold | maintains this garbage information in a garbage information database beforehand.

  Further, in the second embodiment, the parameter is set based on the number of pedestrians detected from the infrared image of the previous frame. However, not only the previous frame but also a plurality of frames from the previous frame to several frames before. A parameter may be set based on the number of pedestrians detected. For example, when the number of pedestrians detected is greater than the number of people determined in succession for several frames, the parameter for people is set.

  In the first embodiment, parameters are set using past crowd information, and in the second embodiment, parameters are set using the number of pedestrians detected in the previous frame. The parameter may be set by using both the past crowd information and the number of pedestrians detected in the previous frame. For example, even if no pedestrians are detected in past crowd information, if more than the number of crowds detected in the previous frame is detected, it is highly likely that they will be detected after the current frame. To do. In addition, if pedestrians more than the number of crowds are detected in the past crowd information and no pedestrians greater than the number of crowds are detected in the previous frame, even if the parameter for crowds is set, it will be detected after the current frame. Because there is a low possibility that

It is a lineblock diagram of the night view system concerning a 1st embodiment. It is an example of the image displayed on the display apparatus of the night view system which concerns on this Embodiment, (a) is a case where there is no detection failure, and (b) is a case where there is a detection failure. It is a flowchart which shows the flow of a process in ECU of the night view system which concerns on 1st Embodiment. It is a block diagram of the night view system which concerns on 2nd Embodiment. It is a flowchart which shows the flow of a process in ECU of the night view system which concerns on 2nd Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1,11 ... Night view system, 2 ... Infrared projector, 3 ... Infrared camera, 4 ... Display apparatus, 5, 15 ... ECU, 5a, 15a ... Parameter adjustment part, 5b, 15b ... Pedestrian detection part, 5c ... Detection result Analysis unit, 15c ... detection result holding unit, 6 ... navigation system, 6a ... current position detection unit, 6b ... map / environment information database, 6c ... crowd information database

Claims (5)

An object detection device for detecting an object from a captured image captured by an imaging means,
Detection reference setting means for setting a detection reference based on the number of detection target objects;
An object detection unit that detects a detection target object from a captured image based on the detection criterion set by the detection criterion setting unit;
The detection reference setting unit sets a relaxed detection reference when the number of detection target objects is large compared to when the number is small.   The object detection apparatus according to claim 1, wherein the detection reference setting unit sets a more relaxed detection reference as the number of detection target objects is larger. Detection information holding means for holding at least number information of the detection target object and location information where the detection target object is detected in association with each other;
The object detection apparatus according to claim 1, wherein the detection reference setting unit sets a detection reference with reference to information held by the detection information holding unit. The object detection means determines that the object to be detected is an object to be detected when an evaluation value when performing object detection exceeds a detection criterion;
The detection reference setting means sets a smaller value as a detection reference when the number of detection target objects is large than when it is small. Object detection device. The object detection means determines that the object to be detected is an object to be detected when an evaluation value when performing object detection is lower than a detection reference;
The detection reference setting unit sets a larger value as a detection reference when the number of detection target objects is large than when the number is small. Object detection device.

Images