JP4888707B2 - Suspicious person detection device - Google Patents

Description

  The present invention relates to a suspicious person detection device that recognizes a suspicious person outside a vehicle based on a captured image.

  In order to prevent vehicle theft and damage to the vehicle, it is important to detect suspicious persons who perform suspicious behavior around the vehicle. In the conventional technology, when a user leaves the vehicle and enters a warning mode, an object approaching the vehicle from the periphery of the vehicle is monitored using a surrounding sensor. When an approaching object is detected, a threatening alarm is given to the object, and at the same time, a surrounding image of the vehicle is transmitted to the information center, and the surrounding information is accumulated in the information center (see, for example, Patent Document 1). This apparatus includes a peripheral sensor that radiates a radio wave of a predetermined frequency band (for example, a GHz band) that can be reflected by an object including a human and receives the reflected wave that is reflected when the radio wave hits the object. Thereby, the intrusion of the object into the detection area, that is, the approach of the object to the vehicle can be detected. An imaging apparatus that acquires a surrounding image includes one omnidirectional camera or a plurality of cameras having a wide viewing angle. In this apparatus, when the surrounding sensor detects the approach of the object to the vehicle, the captured image around the vehicle is only transmitted to the information center, and the captured image is not evaluated. Therefore, it is impossible to determine whether the vehicle approaching object is captured in the captured image or whether the vehicle approaching object is a person on the spot. Therefore, there is a high possibility of misidentifying suspicious behavior on the vehicle.

  Separately from this, as a vehicle monitoring system for evaluating a captured image for monitoring, motion detection is performed from an image captured by using an imaging means installed in the vehicle interior through the vehicle interior and window. Based on the results, it is known that the presence of an intruder or a suspicious person is determined and the determination result is notified by a notification means (see, for example, Patent Document 2). In this vehicle monitoring system, a background difference image that is a difference image between an input image input from an imaging unit installed on a ceiling portion of the vehicle and a preset background image is binarized. As a result, a binarized image including a target area that changes with respect to the background image and a background area that does not change is generated. Here, the background image is an image captured in a state where there is no moving object within the imaging range. By comparing the binarized images, an overlapping area that remains unchanged in the target area, an increased area that has changed from the background area to the target area, and a reduced area that has changed from the target area to the background area are extracted. Two-dimensional coordinates to represent the number of pixels belonging to the overlapping area and the integrated area obtained by integrating the increasing area and decreasing area adjacent to the overlapping area, and the increasing area and decreasing area, and the pixel position. A region parameter composed of an X coordinate sum and a Y coordinate sum obtained by adding the coordinate values of the X coordinate and the Y coordinate for each pixel for all pixels belonging to the region is obtained. And based on this area | region parameter, the motion of the moving object shown by the integrated area | region is detected for every integrated area | region. In this system, since the captured image acquired by the imaging unit is image-evaluated, the determination accuracy of the suspicious object is high, but the entire input omnidirectional captured image is the object of image evaluation, so the computational load of the image evaluation process Is disadvantageous.

JP 2003-157383 A (paragraph numbers 0042 and 0042, abstract) JP 2006-59252 A (paragraph numbers 0012-0035)

  In view of the above situation, an object of the present invention is to provide a suspicious person detection device that efficiently evaluates an acquired captured image around a vehicle so as not to increase a calculation load and recognizes a suspicious person outside the vehicle. It is to be.

In order to solve the above problems, a suspicious person detection device according to the present invention for recognizing a suspicious person outside a vehicle based on a photographed image is installed in a vehicle and obtains a photographed image around the vehicle through a window. When an image cut section for outputting an image area of each obtained through a plurality of windows from said photographing field image isolate as the window area image, based on the proximity to the window of the object outside the vehicle, it was the cut An image selection unit that selects an evaluation target image from a plurality of window region images, and determines whether an approaching object is a suspicious person based on the evaluation target image.

  As in the above feature configuration, by evaluating the window area image related to the object approach signal among the plurality of window area images as the evaluation target image, the window area image that does not capture the approaching object is determined from the evaluation target image. Therefore, the calculation load is reduced and the image evaluation is performed efficiently.

  Further, as a second characteristic configuration in the present invention, an object approach determination unit that outputs an object approach signal based on an output signal from a sensor unit provided in the vehicle to detect an object approaching the vehicle, A photographing control unit that activates the photographing unit based on an object approach signal. With such a configuration, the photographing unit can be activated only when an object approaches, so that battery consumption can be suppressed.

  Further, when the approaching object moves around the vehicle body, the approaching object cannot be captured by the same window area image. In this case, a new window area image is set as an evaluation target image by an object approach signal from the output of another sensor unit that has detected an approaching object. However, in order to accurately follow and monitor the approaching object, it is preferable that the adjacent window area image is set as the evaluation target image before the approaching object captured in one window area image moves to the other window area image. . For this reason, the present invention includes, as a third feature configuration, an evaluation image adding unit that instructs the image selection unit to select a window region image to be further evaluated based on the movement of the object. .

  When determining whether an object approaching the vehicle body is a suspicious person, it is necessary to confirm that the object is a person. Therefore, the present invention can include a face detection unit that determines that a part of the object is a human face as a fourth characteristic configuration. For this face detection, for example, a face detection algorithm that analyzes pixel values of a region where the contour is detected and detects a region that is regarded as a face in the image is used. As a result, the face detection area data consisting of the face position (coordinate position in the image) and the face size (vertical and horizontal size of the rectangular pixel area based on the face position), the face orientation, and the reliability calculated at the time of face detection are used for face detection. Output as information.

  Further, in the present invention, as a fifth feature configuration, when determining whether or not the person determined from the detection of the face is a suspicious person, an action of looking into the vehicle while placing a hand around the face is regarded as a suspicious action . Image evaluation of whether or not a hand is present around the face is performed by checking that an area having a pixel value similar to the skin area of the face exists in the vicinity of the face, or by detecting the contour of the face peripheral area. Check if the line is similar to the hand shape.

  In the present invention, as a sixth feature configuration, it is determined whether or not the person is a suspicious person when the face width of the person determined by the face detection unit is longer than a predetermined length. With such a configuration, since it can be determined that the person is a suspicious person when the person approaches sufficiently, it is possible to reliably determine whether or not the person is a suspicious person.

  Furthermore, in the present invention, as a seventh characteristic configuration, a peeping determination unit that determines peeping of the window is provided, and the peeping determination unit is configured such that the person determined by the face detection unit has a window area longer than a predetermined time. If it exists in the image, it is determined that it is a peep. With such a configuration, it can be determined whether or not the approaching person is a suspicious person.

  Further, according to the present invention, as an eighth characteristic configuration, when the approaching object is captured over a plurality of window region images, the approaching object is regarded as wandering around the vehicle. Thereby, it can be determined with high accuracy that the approaching object is a suspicious person.

  Furthermore, this invention can be provided with the alerting | reporting part which alert | reports a suspicious person detection as a 9th characteristic structure, when a suspicious person is detected. With this configuration, a suspicious person approaching the vehicle can be threatened, and vehicle theft and damage to the vehicle can be prevented in advance. As such a notification method, various forms such as visual threat and auditory threat can be adopted.

Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 is a block diagram schematically showing a configuration of a suspicious person detection device according to the present invention. The suspicious person detection device includes the photographing means 1 and the sensor unit 2 and detects a suspicious person. The sensor unit 2 includes obstacle sensors 31 to 35 such as an infrared sensor, and outputs a signal in response to detection of an object approaching the vehicle. Based on this output signal, the object approach determination unit 3 determines whether an object is approaching the vehicle. If the object is approaching, the imaging control unit 4 activates the imaging unit 1 and outputs a control signal for starting imaging to the imaging unit 1. The photographing means 1 starts photographing based on this control signal. The photographing means 1 is composed of a monitoring camera 21 made up of a photographing lens, a photoelectric conversion element, and the like. However, in the present invention, unless specifically distinguished, the photographed image and image data obtained by photographing are collectively referred to as “captured image”. In addition, this suspicious person detection device has a function unit for performing various operations related to detection of a suspicious person using a CPU as a core member, by hardware and / or software.

  FIG. 2 is a diagram illustrating an outline of the arrangement of the monitoring camera 21 and the obstacle sensors 31 to 35, and FIG. 3 is a diagram illustrating a monitoring area of the obstacle sensors 31 to 35. As shown in FIG. 2, the monitoring camera 21 is arranged in the vicinity of the windshield so as to face the rear of the vehicle, and the inside of the vehicle is photographed without changing the direction of the monitoring camera 21 and the front glass on the driver's seat side, The area around the vehicle is photographed through the glass, rear glass, rear glass on the passenger seat side, and front glass on the passenger seat side. Also, the obstacle sensors 31 to 35 are disposed in the vicinity of each door panel, and predetermined monitoring areas 41 to 45 are determined as shown in FIG. Therefore, it is possible to detect an object that enters the monitoring areas 41 to 45 based on the obstacle sensors 31 to 35.

  The captured image acquired by the imaging unit 1 is stored in the memory 5. The captured image stored in the memory 5 is segmented into a plurality of window area images W1 to W5 that are constituted by the window portion of the vehicle in the captured image by the image segmenting unit 7. Since the divided window area images W1 to W5 are images originally captured by one monitoring camera 21 provided near the windshield as described above, the images are captured from an oblique direction that is not perpendicular to the window direction. Some include images and have distortion. Therefore, the distortion correction unit 8 performs distortion correction on the divided window area images W1 to W5.

  FIG. 4A is a diagram illustrating an example of an image photographed by the photographing unit 1. Since the surveillance camera 21 is arranged from the vicinity of the windshield to the rear of the vehicle, the photographed image includes the driver's seat and front passenger seat, the driver's seat side front glass (monitoring window 51), and the driver's seat side rear glass ( The monitoring window 52), the rear glass (monitoring window 53), the rear glass on the passenger seat (monitoring window 54), and the front glass on the passenger seat (monitoring window 55) are shown. Next, as shown in FIG. 4B, the above-described image segmentation unit 7 segments each window as window region images W1 to W5. The cut window area images W1 to W5 are distorted using a known image processing technique so that the image is the same as the image photographed from the front as shown in FIG. Corrected images X1 to X5 are produced. Therefore, an image similar to the image captured for each of the monitoring windows 51 to 55 can be obtained by the monitoring camera 21 provided near the windshield.

  Based on the images X1 to X5 whose distortion has been corrected by the distortion correction unit 8 and the outputs of the obstacle sensors 31 to 35, an image determined by the image selection unit 9 that an object approaching the vehicle is shown. Selected. The suspicious person determination means 10 determines whether or not the object shown in the selected image is a suspicious person. As a result, the window region image that does not appear to capture the approaching object is temporarily removed from the evaluation target image, so that the arithmetic processing can be reduced and the image evaluation can be performed efficiently. When the suspicious person determination unit 10 determines that the object approaching the vehicle is a suspicious person, the notification unit 15 performs notification using a buzzer, a lamp, or the like. Therefore, it is possible to warn a suspicious person by this notification and to notify a person other than the suspicious person around the vehicle that there is a suspicious person. This can be prevented beforehand.

  The suspicious person determination unit 10 determines whether or not the object shown in the image selected by the image selection unit 9 is a suspicious person as described above, and the face detection unit 11 and the evaluation image addition unit are included as components thereof. 12, a peep determination unit 13 and a wrinkle determination unit 14 are provided. The face detection unit 11 detects a face part from an object shown in the image selected by the image selection unit 9. Therefore, it is possible to determine whether or not the object is a person by this detection. The peeping determination unit 13 determines that it is a suspicious person who is peeping into the vehicle when a hand is detected around the face detected by the face detection unit 11. As image evaluation at this time, it is checked that an area having a pixel value similar to the skin area of the face exists in the vicinity of the face, or the outline detected in the peripheral area of the face is the shape of the hand. Or check if it is similar to. When the suspicious person moves to another window, the evaluation image adding unit 12 selects the window region images W1 to W5 to be further evaluated based on the movement to the image selecting unit 9. Command. Therefore, before the suspicious person moves toward the adjacent window area image, the adjacent window area image can be set as the evaluation target image, and therefore follow-up monitoring can be performed. The heel determining unit 14 hesitates that an object approaching the vehicle is moving around the vehicle for a predetermined period across a plurality of window region images including the window region image added by the evaluation image adding unit 12. It is determined that the person is a suspicious person. As described above, when it is determined that the object is a suspicious person, the image recording unit 6 records a photographed image photographed by the photographing unit 1 via the memory 5.

  Next, a method for determining whether or not an object shown in the distortion-corrected images X1 to X5 is a suspicious person who is looking into the vehicle will be described. FIG. 5 is a diagram schematically illustrating the face detection process in the face detection unit 11. First, feature points are extracted with respect to an object shown in the distortion-corrected images X1 to X5. Examples of this feature point include the width and height of the face, left-right symmetry of eyes, ears, eyebrows, and the like, and the position of the nose and mouth in the recognized face. When the object approaching the vehicle is a person based on the extracted features, it is determined whether it is a suspicious person looking into the vehicle from the movement of the person's face or a person who is just passing by. Becomes easy.

  As described above, the evaluation image adding unit 12 is configured to follow the suspicious person when the suspicious person who is looking into the vehicle moves from the window that was looking into the window to the next window. The image selection unit 9 is commanded to select W5. FIG. 6 is a diagram illustrating an example in which an image is selected by capturing the movement of the suspicious person's face in this case. When the suspicious person shown in FIG. 6A moves within the same image as shown in FIG. 6B, the movement of the suspicious person can be known by the change in the output signals of the obstacle sensors 31-35. Furthermore, although the details are omitted, the speed vector of the suspicious person's face between two frames is obtained by using the known optical flow technique, and the direction of the suspicious person's movement is known by reading this speed vector. You can also. Even when the movement is to another window as shown in FIG. 6 (b) to FIG. 6 (c), the movement can be recognized by the output signals of the obstacle sensors 31 to 35 and the optical flow technique described above. Based on the above, the evaluation image adding unit 13 instructs the image selecting unit 9 to select the window area images W1 to W5 in order to follow and monitor the suspicious person.

  When looking into the interior of the vehicle, as shown in FIG. Therefore, if the peeping determination unit 13 recognizes such a posture as a suspicious action and recognizes such a posture, it can be determined that the suspicious person is looking into the vehicle.

  Next, the present embodiment will be described using a flowchart. FIG. 8 is a flowchart for detecting a suspicious person. First, the obstacle sensors 31 to 35 included in the sensor unit 2 start detecting an object approaching the vehicle (step # 01). If the object falls within a predetermined range set in advance by the user (step # 02: Yes), the object approach determination unit 3 takes a signal indicating the approach of the object based on the output signals of the obstacle sensors 31 to 35. It is output to the control unit 4. In response to this signal, the imaging control unit 4 outputs a control signal to the imaging means 1. The photographing means 1 is activated based on this control signal and starts photographing (step # 03). On the other hand, if there is no approach of the object within the predetermined range (step # 02: No), the photographing means 1 is not activated, so no photographing is performed, and an object approaching the vehicle is continuously detected.

  When the photographing means 1 is activated, the monitoring camera 21 disposed so as to face the rear of the vehicle near the windshield shoots an object approaching the vehicle. The acquired photographed image is stored in the memory 5, and is segmented as window area images W1 to W5 by the image segmenting unit 7 (step # 04). Since these window area images W1 to W5 are images taken with the monitoring camera 21 facing rearward in the vehicle, the window area images that are not in the front direction (rear glass direction) from the monitoring camera 21 (front glass on the driver's seat side, driver side) Some of the rear glass, the rear glass on the passenger seat side, and the front glass on the passenger seat side have distortion like an image viewed from an oblique direction that is not perpendicular to the window direction. This distortion is corrected by the distortion correction unit 8 using a known image processing method (step # 05), and images X1 to X5 having distortion correction such as an image photographed from a direction perpendicular to the window direction are produced. Is done. In addition to this distortion correction, brightness and contrast are also adjusted for each segmented image.

  Based on the distortion-corrected images X1 to X5 and the detection by the obstacle sensors 31 to 35, an image showing an approaching object is specified (step # 06), and the image selection unit 9 selects the image. The Next, the suspicious person determination means 10 determines whether or not the approaching object shown in the specified image is a suspicious person through face detection (step # 07). Here, the determination routine in the suspicious person determination means 10 will be described later. When the suspicious person judging means 10 judges that the object shown in the image is a suspicious person (step # 08: Yes), the image recording unit 6 shows the suspicious person photographed by the photographing means 1 via the memory 5. A captured image is recorded (step # 11). Further, the notification unit 15 notifies that there is a suspicious person around the vehicle (step # 12). This notification allows not only a warning to a suspicious person but also to notify a person other than the suspicious person around the vehicle that there is a suspicious person.

  On the other hand, if it is determined by the suspicious person determination means 10 that the object shown in the image is not a suspicious person (step # 08: No), if the notification is already in progress (step # 09: Yes), the notification unit 15 notifies the user. The photographing unit 1 finishes photographing (step # 10). If the notification is not in progress (step # 09: No), the photographing means 1 ends the photographing (step # 10).

  Next, determination of a suspicious person by the suspicious person determination means 10 will be described using a flowchart. FIG. 9 is a flowchart for determining a suspicious person. In step # 06 in FIG. 8, an image showing an approaching object is identified based on the distortion corrected images X1 to X5 and detection by the obstacle sensors 31 to 35, and the image selection unit 9 selects the image. Then, the face detection unit 11 included in the suspicious person determination unit 10 performs face detection from the selected image. If no face is shown in the image selected by the image selection unit 9 (step # 21: No), the suspicious person determining means 10 determines that the object is not a suspicious person (step # 22), and the flow of suspicious person detection is performed. Return.

  If a face is reflected in the image selected by the image selection unit 9 (step # 21: Yes), the peeping determination unit 13 determines whether or not a hand is captured around the face. If a hand is shown around the face (step # 23: Yes), the suspicious person determination means 10 determines that the object is a suspicious person (step # 26), and returns to the suspicious person detection flow. If the hand is not captured around the face (step # 23: No), the eyelid determining unit 14 determines whether or not the object is wandering around the vehicle. If the object is wandering around the vehicle (step # 26: Yes), the suspicious person determination means 10 determines that the object is a suspicious person (step # 26) and returns to the suspicious person detection flow. If the object does not hesitate around the vehicle (step # 24: No), it is determined whether the object is around the vehicle for a predetermined period. If there is an object around the vehicle for a predetermined time (step # 25: Yes), the suspicious person determination means 10 determines that the object is a suspicious person (step # 26), and returns to the suspicious person detection flow. On the other hand, if there is no object around the vehicle for a predetermined period (step # 25: No), the suspicious person determination means 10 determines that the object is not a suspicious person (step # 22), and returns to the suspicious person detection flow. In this way, the suspicious person determination means 10 can determine whether or not the object approaching the vehicle is a suspicious person based on the image selected by the image selection unit 9.

Other Examples In the above embodiment, the monitoring camera 21 captures an object approaching the vehicle in the suspicious person detection device, but the present invention is not limited to this. Of course, it is possible to use the camera as a camera provided in a device for photographing the driver and preventing the driver from falling asleep while the vehicle is running.

  In the above embodiment, the obstacle sensors 31 to 35 are described as being configured by infrared sensors, but the present invention is not limited thereto. Obviously, the obstacle sensors 31 to 35 can be configured as long as they can detect an object approaching the vehicle.

  In the above-described embodiment, the monitoring unit 21 provided in the photographing unit 1 is described as being one, but the present invention is not limited to this. Of course, a configuration including a plurality of surveillance cameras is also possible.

  In the above embodiment, the glass provided in the car is the front glass on the driver's seat side, the rear glass on the driver's seat side, the rear glass, the rear glass on the passenger seat side, and the front glass on the passenger seat side, but is not limited thereto. If the vehicle has at least one glass, it is naturally possible to configure the suspicious detection device according to the present invention, and the suspicious person detection device according to the present invention is also provided with a glass other than the above. It is of course possible to construct

  In the above embodiment, as examples of feature points used when detecting a face from the distortion-corrected images X1 to X5, the horizontal and vertical widths of the face, the symmetry of eyes, ears, eyebrows and the like, and further recognition Although described as the position of the nose and mouth in the face made, it is not limited to this. For example, it is naturally possible to configure a person such as a mask or hat at the mouth that covers the face deeply to be recognized as a suspicious person, and it is also possible to set other feature points. is there.

  In the embodiment described above, whether or not the object is on the periphery of the vehicle at the time of determining the suspicious person and whether or not the object is in the vicinity of the vehicle for a predetermined time are determined by the image selected by the image selection unit 9. However, the present invention is not limited to this. Of course, it is also possible to make a determination based on the output signal of the sensor provided in the sensor unit 2.

  In the above-described embodiment, it has been described that an approaching object is detected by the obstacle sensor, but the present invention is not limited to this. Of course, it is also possible to detect the approach of the object by the photographing means. A known method is used as means for detecting the approach of the object by the approaching means. For example, it is detected based on the difference between the immediately preceding vehicle surrounding image and the current vehicle surrounding image or by the optical flow technique.

Block diagram schematically showing the configuration of the suspicious person detection device Diagram showing the outline of the arrangement of surveillance cameras and obstacle sensors Diagram showing the obstacle sensor monitoring area Diagram showing the process of creating a distortion-corrected image from a captured image Diagram showing the outline of the face detection process Figure for selecting an image as a suspicious person moves The figure which shows an example of the attitude | position which performs peeping determination The figure which shows the flowchart for detecting a suspicious person The figure which shows the flowchart for determining a suspicious person

Explanation of symbols

1: Shooting means 2: Sensor unit 3: Object approach determination unit 4: Shooting control unit 7: Image segmentation unit 8: Distortion correction unit 9: Image selection unit 10: Suspicious person determination unit 11: Face detection unit 12: Addition of evaluation image Unit 13: Peep determination unit 15: Notification unit 21: Monitoring cameras 31 to 35: Obstacle sensors 41 to 45: Monitoring areas 51 to 55: Monitoring windows W1 to W5: Window region images X1 to X5: Distorted image

Claims (9)

In the suspicious person detection device that recognizes the suspicious person outside the vehicle based on the captured image,
And imaging means for obtaining captured image around the vehicle through the window while being installed in a vehicle, an image segmentation unit for outputting the image area of each obtained through a plurality of windows from said photographing field image isolate as the window region image An image selection unit that selects an evaluation target image from the plurality of divided window region images based on the approach of the object outside the vehicle to the window, and the object approaching based on the evaluation target image is suspicious A suspicious person detection device characterized by determining whether or not a person is a person.   An object approach determination unit that outputs an object approach signal based on an output signal from a sensor unit provided in the vehicle to detect an object approaching the vehicle, and a photographing control that activates the photographing unit based on the object approach signal The suspicious person detection device according to claim 1 provided with a section.   3. The suspicious person detection device according to claim 1, further comprising an evaluation image addition unit that instructs the image selection unit to select a window region image to be a further evaluation target image based on the movement of the object.   The suspicious person detection device according to any one of claims 1 to 3, further comprising a face detection unit that determines that a part of the object is a human face.   The suspicious person detection device according to claim 4, wherein a suspicious person is determined to be a suspicious person when a hand is detected around the face detected by the face detection unit.   The suspicious person detection device according to claim 4, wherein the suspicious person detection device determines whether or not the person is a suspicious person when a face width of the person determined by the face detection unit is longer than a predetermined length.   A peeping determination unit for determining peeping in the window, wherein the peeping determination unit determines that the person is determined to be peeping when the person determined by the face detection unit is present in the window area image for longer than a predetermined time. The suspicious person detection device according to claim 4.   The suspicious person detection device according to claim 1, wherein the suspicious person detection device determines that the object is a suspicious person when it is detected that the object moves over a plurality of window regions for a predetermined period.   The suspicious person detection device according to any one of claims 1 to 8, further comprising a notification unit that notifies suspicious person detection when a suspicious person is detected.

Images