JP2014146164A - Object detection apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an object detection apparatus which detects an object with high accuracy even if there is a high-luminance area other than the object.SOLUTION: An object detection apparatus 1 for detecting an object (a pedestrian, for example) from an image includes: imaging means 10 for capturing images in different exposure conditions; comparison means 26, 27 which compare characteristics of a shape of a high-luminance area (a position of the center of gravity of the high-luminance area, and the number of high-luminance areas, for example) between object candidate areas located in the same image position and extracted from a plurality of images captured by the imaging means 10 in different exposure conditions; and determination means 28, 29 which determine whether the object candidate area is the object or not, on the basis of a comparison result of the comparison means. The determination means determine that the object candidate area is a light-emitting object when characteristic change in the shape of the high-luminance area is small between the object candidate areas with different exposure conditions, on the basis of the comparison result of the comparison means.

Description

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

  The driver of the vehicle must pay attention to pedestrians and the like existing around the vehicle while traveling. In order to reduce such a burden on the driver, an apparatus for detecting a pedestrian around the vehicle has been developed. For example, Patent Document 1 discloses that a high brightness area is extracted as a candidate area for a pedestrian from an image captured by a far-infrared camera, and a pedestrian is detected using the candidate area.

JP 2008-65757 A

  When detecting pedestrians based on high-luminance areas in the image, illuminants such as night street lights and store lighting are also high-luminance, so if the shape of the illuminant resembles a pedestrian, walk around the illuminant. There is a possibility of false detection as a person.

  Therefore, an object of the present invention is to provide an object detection device that can detect an object with high accuracy even when there is a high-luminance region other than the object.

  An object detection apparatus according to the present invention is an object detection apparatus that detects an object from an image, and is extracted from an imaging unit that can capture images under different exposure conditions and a plurality of images that are captured by the imaging unit and that have different exposure conditions. Comparison means for comparing features of the shape of the high-intensity area between object candidate areas at the same image position, and determination means for determining whether the object candidate area is an object based on the comparison result of the comparison means; It is characterized by providing.

  In this object detection apparatus, the imaging means captures images under a plurality of different exposure conditions, and acquires a plurality of images with different exposure conditions. Examples of exposure conditions include exposure time (shutter speed) and lens aperture. If the exposure conditions are different, the amount of light detected by the imaging means changes even when the same object is imaged. The object detection device extracts object candidate regions at the same image position from a plurality of images with different exposure conditions, extracts a high-luminance region from a plurality of object candidate regions with different exposure conditions, and Compare the characteristics of the shape of the high brightness area. In the case of a luminescent material (non-object) such as illumination, the light is emitted by itself and diffused radially from the light emission center to become bright. Therefore, even if the exposure condition is changed, the region extracted as the high luminance region is hardly changed, and the change in the shape of the high luminance region is small. On the other hand, in the case of an object such as a pedestrian, light such as illumination is reflected and brightened. Therefore, when the exposure condition is changed, the region extracted as the high luminance region changes, and the shape of the high luminance region changes greatly. For example, when the exposure time is shortened, a portion where the amount of reflected light is small does not have high luminance, and the high luminance region becomes small and is eventually divided. Therefore, the object detection apparatus can determine whether or not the object candidate area is an object based on the comparison result of the characteristics of the shape of the high luminance area between the plurality of object candidate areas with different exposure conditions. As described above, in the object detection device, even if there is a high-luminance region other than the target object by comparing the shape characteristics of the high-luminance region between the target object candidate regions extracted from the images of the plurality of exposure conditions. The object can be detected with high accuracy.

  In the object detection apparatus of the present invention, the determination means emits light from the object candidate area when the change in the feature of the shape of the high brightness area is small between the plurality of object candidate areas having different exposure conditions from the comparison result of the comparison means. Judge that it is a thing.

  As described above, in the case of a illuminant such as illumination, even if the exposure condition is changed, the region extracted as the high luminance region is hardly changed, and the change in the shape of the high luminance region is small. Therefore, in the object detection device, when the feature characteristics of the high brightness area are compared between a plurality of target object areas having different exposure conditions, and the change in the shape characteristics of the high brightness area is small, the object candidate area Can be determined to be a luminescent material. As a result, the object candidate area | region which was not determined as a light-emitting object can be determined as a target object.

  In the object detection apparatus of the present invention, the comparison means compares the barycentric positions of the high-luminance areas between a plurality of object candidate areas having different exposure conditions.

  In the case of a illuminant such as illumination, the brightness diffuses radially from the light emission center, and the diffused region is extracted as a high luminance region. Accordingly, since the emission center always has the highest luminance, the center of gravity position of the high luminance region does not change (or changes are small) even if the exposure conditions are changed. On the other hand, in the case of an object such as a pedestrian, the light is reflected and brightened, and the reflected area is extracted as a high luminance area. Accordingly, since the portion with a larger amount of reflected light has higher luminance, changing the exposure condition also changes the position of the center of gravity of the high luminance region. Therefore, in the object detection device, it is determined with high accuracy whether or not the feature of the shape of the high luminance region has changed by comparing the center of gravity position of the high luminance region between a plurality of candidate object regions having different exposure conditions. it can.

  In the object detection apparatus of the present invention, the comparison means compares the number of high-luminance areas between a plurality of object candidate areas having different exposure conditions.

  In the case of illuminants such as illuminations, even if the exposure conditions are changed as described above, the change in the shape of the high-brightness area is small, so the high-brightness area hardly divides and the number of high-brightness areas also changes little. do not do. On the other hand, in the case of an object such as a pedestrian, if the exposure condition is changed as described above, the change in the shape of the high luminance region is large. With this change in the shape of the high-luminance region, the shape of the region becomes small, and the region may eventually be divided. Therefore, in the object detection device, it is possible to determine with high accuracy whether or not the feature of the shape of the high-luminance region has changed by comparing the number of high-luminance regions between a plurality of candidate object regions having different exposure conditions. .

  According to the present invention, even if there is a high-luminance region other than the target object by comparing the characteristics of the shape of the high-luminance region between the target object candidate regions extracted from the images of the plurality of exposure conditions, the target object Can be detected with high accuracy.

It is a block diagram of the pedestrian detection apparatus which concerns on this Embodiment. It is an example of a processing process in the case of being determined as a pedestrian, (a) is a pedestrian candidate area extracted from an image having a long exposure time, and (b) is a pedestrian candidate area in an image having a long exposure time (C) is a high brightness area extracted from a pedestrian candidate area in an image with an intermediate exposure time, and (d) is a pedestrian candidate in an image with a short exposure time. This is a high luminance area extracted from the area. It is an example of a processing process when it is determined as a luminescent object, (a) is a pedestrian candidate area extracted from an image with a long exposure time, and (b) is a pedestrian candidate area with an image with a long exposure time. (C) is a high brightness area extracted from a pedestrian candidate area in an image with an intermediate exposure time, and (d) is a pedestrian candidate in an image with a short exposure time. This is a high luminance area extracted from the area. It is a flowchart which shows the flow of the pedestrian detection process in ECU of FIG.

  Hereinafter, an embodiment of an object detection device according to the present invention will be described with reference to the drawings. In addition, the same code | symbol is attached | subjected about the element which is the same or it corresponds in each figure, and the overlapping description is abbreviate | omitted.

  In the present embodiment, the object detection device according to the present invention is applied to a pedestrian detection device mounted on a vehicle. The pedestrian detection device according to the present embodiment detects a pedestrian from an image captured by a visible light camera, and provides information on the detected pedestrian to various driving support devices. In particular, in the present embodiment, a pedestrian detection process at night (which may be applied when the surroundings in a tunnel or the like are dark in the daytime) in the pedestrian detection apparatus will be described. In addition, about the pedestrian detection process in daytime, you may apply the pedestrian detection method demonstrated in this Embodiment, However, What is necessary is just to apply the pedestrian detection method suitable for the other known daytime.

  With reference to FIGS. 1-3, the pedestrian detection apparatus 1 which concerns on this Embodiment is demonstrated. FIG. 1 is a configuration diagram of a pedestrian detection device according to the present embodiment. FIG. 2 is an example of a processing process when it is determined that the person is a pedestrian. FIG. 3 is an example of a processing process in the case of being determined as a luminescent material.

  The pedestrian detection device 1 extracts a pedestrian candidate area from the image, and detects a pedestrian based on the high brightness area in the extracted pedestrian candidate area. In particular, the pedestrian detection apparatus 1 has the same image position respectively extracted from images of three different exposure conditions (exposure times) so as not to erroneously detect luminous objects such as night street lamps and store lighting as pedestrians. It is determined whether the pedestrian candidate area is a luminescent object or a non-luminescent object by comparing the characteristics of the shape of the high brightness area between the pedestrian candidate areas. The pedestrian detection device 1 includes a visible light camera 10 and an ECU [Electronic Control Unit] 20.

  The visible light camera 10 is a visible light camera that images the front of the host vehicle using a CCD (Charge Coupled Device), a CMOS (Complementary Metal Oxide Semiconductor), or the like. The visible light camera 10 is attached to a predetermined height position in the center of the front side of the host vehicle. The visible light camera 10 is a camera that can capture images by changing exposure conditions. An example of the exposure condition is an exposure time (shutter speed). The visible light camera 10 may be a color camera or a monochrome camera, as long as at least luminance information can be obtained. The visible light camera 10 images the front of the host vehicle at regular time intervals, and transmits the captured image data to the ECU 20 as an image signal. In the present embodiment, the visible light camera 10 corresponds to the imaging means described in the claims.

  The ECU 20 is an electronic control unit including a CPU [Central Processing Unit], a ROM [Read Only Memory], a RAM [Random Access Memory], and the like. The ECU 20 includes a pedestrian model database 21 in a predetermined storage area. Further, the ECU 20 loads various programs stored in the ROM into the RAM and executes them by the CPU, whereby the exposure control unit 22, the multiple exposure image acquisition unit 23, the first pedestrian determination unit 24, the first height A luminance region extraction unit 25, a second high luminance region extraction unit 26, a third high luminance region extraction unit 27, a luminescent matter determination unit 28, and a second pedestrian determination unit 29 are configured. In the present embodiment, the second high-intensity region extraction unit 26 and the third high-intensity region extraction unit 27 correspond to the comparison means described in the claims, and the illuminant determination unit 28 and the second pedestrian determination The unit 29 corresponds to a determination unit described in the claims.

  The pedestrian model database 21 is a database that stores a pedestrian model used for pattern matching in the first pedestrian determination unit 24. The pedestrian model database 21 is a database constructed in a predetermined storage area of the ECU 20. Various models are stored as pedestrian models, such as walking pedestrian models, running pedestrian models, stopped pedestrian models, male pedestrian models, female pedestrian models, There are adult pedestrian models, child pedestrian models, elderly pedestrian models, front pedestrian models, back pedestrian models, side pedestrian models, and a combination of these pedestrian models. There is also.

  The exposure control unit 22 is a processing unit that controls the imaging of the visible light camera 10 by changing exposure conditions (particularly, exposure time). The exposure control unit 22 is connected to the visible light camera 10 via a bus such as IEEE1394 in order to perform bidirectional communication with the visible light camera 10. The exposure control unit 22 controls the exposure of the visible light camera 10 at regular intervals, and images each of three different exposure times (long exposure time> intermediate exposure time> short exposure time). For these three types of exposure times, a long exposure time is an exposure time that can sufficiently capture a pedestrian in an urban area at night, and an intermediate exposure time and a short exposure time are sequentially set as a short exposure time. Even with a short exposure time, the exposure time is such that light from a luminescent material such as a street lamp at night or store lighting can be sufficiently imaged. The longer the exposure time, the brighter the image.

  The multiple exposure image acquisition unit 23 is a processing unit that acquires an image captured by the visible light camera 10 whose exposure is controlled by the exposure control unit 22. In the multiple exposure image acquisition unit 23, a long exposure time (brightest) image, an intermediate exposure time (intermediate brightness) image, and a short exposure time (longest) captured by the visible light camera 10 at regular time intervals. Acquire a dark image.

  The first pedestrian determination unit 24 is a processing unit that determines a pedestrian-like region (pedestrian candidate region) from an image with a long exposure time (brightest) acquired by the multiple exposure image acquisition unit 23. This determination method uses a well-known general pedestrian recognition method, for example, pattern matching using a pedestrian model. In the 1st pedestrian determination part 24, if the multiple exposure image acquisition part 23 acquires the image of long exposure time, it will pattern with respect to the image of the long exposure time for every pedestrian model stored in the pedestrian model database 21. Matching is performed, and if there is a matched area, a rectangular pedestrian candidate area is extracted. The pattern matching technique can be realized by the technique disclosed in N. Dalal and B. Triggs. “Histogram of oriented gradient for human detection”. CVPR, 2005, for example.

FIG. 2A shows a pedestrian candidate region C1 _L that is determined as a pedestrian and extracted from an image with a long exposure time (brightest). This pedestrian is exposed to light such as nighttime store lighting, and the light is reflected by white clothes or the like and brightened. In the case of an image with a long exposure time, a pedestrian-like shape can be obtained even with reflected light, and pattern matching using a pedestrian model is possible. FIG. 3A shows a pedestrian candidate region C2 _L that is determined as two luminescent objects likely to be pedestrians and extracted from an image with a long exposure time. These two luminescent materials are diffused radially from the emission center, and are brightened in a circular shape and a rectangular shape. The combination of these two shapes of illuminant has a shape similar to a pedestrian, and has been matched by pattern matching using a pedestrian model.

  The first high-intensity region extraction unit 25 extracts a high-intensity region from the pedestrian candidate region extracted from the image with the long exposure time (brightest) by the first pedestrian determination unit 24, and determines the position of the center of gravity of the high-intensity region. A processing unit for calculating. The first high luminance area extraction unit 25 selects pedestrian candidate areas one by one from the pedestrian candidate areas extracted by the first pedestrian determination unit 24. Then, the first high-brightness region extraction unit 25 extracts a region (high-brightness region) having a luminance higher than the first luminance determination threshold in the selected pedestrian candidate region, and determines the center of gravity position of the extracted region. calculate. The first brightness determination threshold value is a threshold value for determining whether or not an image with a long exposure time has high brightness, and is set in advance based on experiments or the like in consideration of the brightness of the image obtained with a long exposure time. The As a method of this region extraction and center-of-gravity position calculation, for example, for each pixel in the pedestrian candidate region, binarization is performed with a pixel having a luminance higher than the first luminance determination threshold and a pixel equal to or lower than the first luminance determination threshold. By using this method, an area composed of pixels higher than the first luminance determination threshold is labeled, and the barycentric position of the area with the same label is calculated.

FIG. 2B shows a high-intensity region H1 _L1 extracted from the pedestrian candidate region C1 _{L having} a long exposure time (brightest) in FIG. The pedestrians were exposed to light from one side, such as nighttime store lighting, and the one side was particularly bright. Therefore, one side end thereof is extracted as a high-luminance region _{H1 L1,} centroid position _{G1 L1} of the high luminance region _{H1 L1} is calculated. FIG. 3B shows high-intensity regions H2 _L1 and H2 _L2 extracted from the pedestrian candidate region C2 _{L having} a long exposure time shown in FIG. The two illuminants diffused from the emission center and became brighter in a circular shape and a rectangular shape, respectively. Therefore, its circular shape and a rectangular shape is extracted as it is shaped as a high-luminance region _H2 L1, _{H2 L2,} centroid position _{G2 L2} center of gravity _{G2 L1} and the high luminance region _{H2 L2} of the high luminance region _{H2 L1} is calculated It was.

  When the first high-brightness region extraction unit 25 extracts a high-brightness region from the pedestrian candidate region having a long exposure time, the second high-brightness region extraction unit 26 has an intermediate exposure time (intermediate brightness). Whether the center of gravity of the high-intensity area extracted from the same area as the pedestrian candidate area with the long exposure time in the image and the center of gravity of the high-intensity area extracted from within the pedestrian candidate area with the long exposure time are the same position It is a processing part to determine. In the second high-intensity region extraction unit 26, when the processing in the first high-intensity region extraction unit 25 for the selected pedestrian candidate region is completed, the first high-intensity region extraction unit 25 performs a pedestrian candidate with a long exposure time. It is determined whether or not a high-intensity area is extracted from the area, and if it is extracted, a pedestrian candidate area having the long exposure time is extracted from the intermediate exposure time image acquired by the multiple exposure image acquisition unit 23. A region having the same position and the same rectangular shape (pedestrian candidate region having an intermediate exposure time) is extracted. Then, the second high brightness area extraction unit 26 uses the same method as the first high brightness area extraction unit 25 to provide a brightness higher than the second brightness determination threshold in the extracted pedestrian candidate area of the intermediate exposure time. A region having the same is extracted, and the center of gravity position of the extracted region is calculated. The second luminance determination threshold value is a threshold value for determining whether or not the image has a high luminance in an intermediate exposure time (intermediate brightness) image, and an experiment is performed in consideration of the brightness of the image obtained in the intermediate exposure time. Etc., based on the above. The second luminance determination threshold value is smaller than the first luminance determination threshold value because a high-latitude region is extracted from an intermediate brightness image.

  Further, the second high brightness area extraction unit 26 calculates the distance between the center position of the extracted high brightness area and the center position of the high brightness area extracted by the first high brightness area extraction unit 25. The method of determining which centroid position distance is to be calculated is the centroid position of the high brightness area in the area position corresponding to the high brightness area extracted by the first high brightness area extraction unit 25. When there are a plurality of high-luminance regions in the region position, the distance is calculated for each of the plurality of high-luminance regions, and the sum of the plurality of distances is used for the determination. Since the second high-intensity region extraction unit 26 uses an image having an intermediate exposure time (intermediate brightness), the high-intensity region extracted by the second high-intensity region extraction unit 26 has a long exposure time (brightest brightness). ) Is not larger than the high luminance area extracted by the first high luminance area extraction unit 25 using the image of (1). As such, the value of the second luminance determination threshold value is prevented from becoming too small. Then, the second high-intensity area extraction unit 26 determines whether or not the calculated distance is equal to or less than the distance determination threshold. If the distance is equal to or less than the distance determination threshold, the high-intensity area in the pedestrian candidate area with a long exposure time And the center of gravity position of the high brightness area in the pedestrian candidate area having an intermediate exposure time are determined to be the same position. The distance determination threshold value is a threshold value for determining whether or not the center of gravity position of the high luminance area may be regarded as the same position between pedestrian areas of images having different exposure times (different brightness), and is based on experiments or the like. Are preset.

FIG 2 (c), 4 single high luminance pedestrian candidate region C1 _L intermediate exposure time of the same image position is extracted from the pedestrian candidate region C1 _M (Middle brightness) shown in FIG. 2 (a) A region H1 _M1 , a high luminance region H1 _M2 , a high luminance region H1 _M3 , and a high luminance region H1 _M4 are shown. In the case of a pedestrian, since it does not emit light, it is not as bright as the center, and only the portion reflecting light such as nighttime store lighting is brightened. Therefore, as the exposure time is shortened, only the part of the reflected light is detected as a high-brightness area, and the high-brightness area becomes narrower (the center of gravity of the area changes accordingly). The area is divided. Therefore, the high brightness area H1 _L1 with a long exposure time shown in FIG. 2B was divided and extracted as four small high brightness areas H1 _M1 , H1 _M2 , H1 _M3 , and H1 _M4 . In this case, the distance between the high luminance region _{H1 L1} of the center-of-gravity position _{G1 L1} high luminance region _{H1 M1, H1 M2, H1 M3} , H1 respective center of gravity of _{M4 G1 M1, G1 M2, G1} M3, G1 M4 are calculated respectively The sum of the four distances is compared with the distance determination threshold value, and is determined to be larger than the distance determination threshold value, so that the position of the center of gravity changes.

Further, in FIG. 3 (c), FIGS. 3 (a) the pedestrian candidate region C2 _L intermediate exposure time of the same image position pedestrian candidate region C2 _M 2 both extracted from the (brightness of the intermediate) of High luminance regions H2 _M1 and H2 _M2 are shown. In the case of a luminescent material, since the light is emitted, the brightness diffuses radially from the emission center. Therefore, even if the exposure time is shortened, the luminance center is the highest in luminance, so that the position of the center of gravity of the high luminance region does not change and the size of the high luminance region does not become so small. Therefore, the high luminance regions H2 _L1 and H2 _{L2 in} the long exposure time shown in FIG. 3B are slightly reduced and extracted as the high luminance regions H2 _M1 and H2 _M2 . In this case, the barycentric position _{G2 M2} of the center of gravity position _{G2 L2} and the high luminance region _{H2 M2} distance and high luminance area _{H2 L2} between the position of the center of gravity _{G2 L1} of the high luminance region _{H2 L1} and centroid position _{G2 M1} of the high luminance region _{H2 M1} Distances are calculated, each distance is compared with a distance determination threshold value, two distances are determined to be equal to or less than the distance determination threshold value, and the center of gravity position is the same position.

  If the second high-brightness region extraction unit 27 determines that there is a high-brightness region at the same center of gravity, the third high-brightness region extraction unit 27 has a long exposure time for a pedestrian candidate with a short exposure time (darkest). This is a processing unit that determines whether the barycentric position of the high luminance area extracted from the same area as the area and the barycentric position of the high luminance area extracted from the pedestrian candidate area having a long exposure time are the same position. In the third high-intensity region extraction unit 27, when the processing in the second high-intensity region extraction unit 26 for the selected pedestrian candidate region is completed, the second high-intensity region extraction unit 26 performs high-intensity with a long exposure time. It is determined whether the center of gravity of the area and the center of gravity of the high-intensity area at the intermediate exposure time are determined to be the same position. If the center of gravity is determined to be the same, the multiple exposure image acquisition unit 23 An area having the same position and the same rectangular shape as the pedestrian candidate area with a long exposure time (pedestrian candidate area with a short exposure time) is extracted from the acquired image with a short exposure time. Then, the third high brightness area extraction unit 27 has a brightness higher than the third brightness determination threshold in the extracted pedestrian candidate area with a short exposure time by the same method as the first high brightness area extraction unit 25. An area is extracted, and the barycentric position of the extracted area is calculated. The third luminance determination threshold is a threshold for determining whether or not the image has a high luminance in a short exposure time (darkest) image, and is set in advance based on an experiment or the like in consideration of the brightness of the image obtained in a short exposure time. Is done. The third luminance determination threshold value is smaller than the second luminance determination threshold value because a high latitude region is extracted from the darkest image.

  Further, the third high brightness area extraction unit 27 calculates the distance between the center position of the extracted high brightness area and the center position of the high brightness area extracted by the first high brightness area extraction unit 25. Since the third high-intensity area extraction unit 27 uses an image with a short exposure time (darkest), the high-intensity area extracted by the third high-intensity area extraction unit 27 is an image with a long exposure time (brightest). The high luminance region extracted by the first high luminance region extraction unit 25 using the image and the high luminance extracted by the second high luminance region extraction unit 26 using the image of the intermediate exposure time (intermediate brightness). It cannot be larger than the area. To prevent this, the value of the third luminance determination threshold value is prevented from becoming too small. Then, the third high brightness area extraction unit 27 determines whether or not the calculated distance is equal to or less than the distance determination threshold value. If the distance is equal to or less than the distance determination threshold value, the high brightness area in the pedestrian candidate area having a long exposure time is determined. It is determined that the barycentric position of the high-luminance area in the pedestrian candidate area with a short exposure time is the same position. The distance from the center of gravity of the high-brightness region extracted by the first high-brightness region extraction unit 25 is obtained, and the high-brightness region extracted by the second high-brightness region extraction unit 26 is determined based on the distance. The distance from the center of gravity of the image may be obtained, and the determination may be made based on the distance.

FIG. 2D shows one high-intensity region H1 _S1 extracted from the pedestrian candidate region C1 _{S with} the short exposure time (darkest) at the same image position as the pedestrian candidate region C1 _L in FIG. Is shown. In this case, since the exposure time is further shorter than in the case of FIG. 2C, the high brightness areas H1 _M1 , H1 _M3 , and H1 _M4 at the intermediate exposure time shown in FIG. Thus, only the high brightness area H1 _M2 at the intermediate exposure time shown in FIG. 2C is reduced and extracted as the high brightness area H1 _S1 . In this case, the distance is calculated between the position of the center of gravity G1 _S1 between the center of gravity position G1 _L1 of the high luminance region H1 _L1 high luminance region H1 _S1, this distance is compared with the distance determination threshold is determined to be greater than the distance determination threshold, The position of the center of gravity has changed.

Further, in FIG. 3 (d), the pedestrian candidate region C2 _L and short exposure time of the same image position (darkest) two high intensity regions extracted from the pedestrian candidate region C2 _S in FIGS. 3 (a) H2 _S1 and H2 _S2 are shown. In this case, the exposure time is further shorter than in the case of FIG. 3C, but is slightly smaller than the high luminance regions H2 _M1 and H2 _M2 at the intermediate exposure time shown in FIG. , Extracted as high luminance regions H2 _S1 and H2 _S2 . In this case, the barycentric position _{G2 S2} between the center of gravity position _{G2 L2} distance and high luminance area _{H2 L2} between the position of the center of gravity _{G2 L1} of the high luminance region _{H2 L1} and centroid position _{G2 S1} of the high luminance region _{H2 S1} high luminance region _{H2 S2} Distances are calculated, each distance is compared with a distance determination threshold value, two distances are determined to be equal to or less than the distance determination threshold value, and the center of gravity position is the same position.

  Based on the distance determination result in the second high-intensity region extraction unit 26 and the distance determination result in the third high-intensity region extraction unit 27, the illuminant determination unit 28 determines whether the pedestrian candidate region is a luminescent object (non-walking). It is a processing unit that determines whether or not a person is). In the illuminant determining unit 28, when the processing in the third high-intensity region extracting unit 27 for the selected pedestrian candidate region is completed, the second high-intensity region extracting unit 26 in the pedestrian candidate region having a long exposure time. It is determined that the barycentric position of the high brightness area and the barycentric position of the high brightness area in the pedestrian candidate area having an intermediate exposure time are the same position, and the third high brightness area extracting unit 27 determines the pedestrian candidate having a long exposure time. When it is determined that the barycentric position of the high brightness area in the area and the barycentric position of the high brightness area in the pedestrian candidate area having a short exposure time are the same position (that is, three pedestrian candidates having different exposure times) When the barycentric positions of the high-luminance areas are the same between the areas), the pedestrian candidate area is determined as a luminescent material (non-pedestrian).

In the case of the example shown in FIG. 2, as described above, it is determined that the position of the center of gravity of the high-luminance region changes between pedestrian candidate regions C1 _L , C1 _M , and C1 _{S having} different exposure times. On the other hand, in the case of the example shown in FIG. 3, as described above, the barycentric position of the high luminance area is determined to be the same among the pedestrian candidate areas C2 _L , C2 _M , and C2 _{S having} different exposure times. Therefore, the pedestrian candidate area shown in FIG. 3 is determined as a luminescent material (non-pedestrian).

  The second pedestrian determination unit 29 is a processing unit that determines whether or not each pedestrian candidate area is a pedestrian based on the determination result of the light emitting object determination unit 28. In the second pedestrian determination unit 29, when all the extracted pedestrian candidate areas have been determined by the illuminant determination unit 28, the illuminant determination unit 28 determines that each pedestrian candidate area is a luminescent object. A no pedestrian candidate area is determined as a pedestrian. And in the 2nd pedestrian determination part 29, about the pedestrian candidate area | region determined as a pedestrian, pedestrian information, such as a relative position from the own vehicle, is calculated, and the pedestrian detection signal which consists of pedestrian information of each pedestrian Is transmitted to the driving support device.

  With reference to FIG. 1, the flow of operation | movement in the pedestrian detection apparatus 1 is demonstrated. In particular, the flow of processing in the ECU 20 will be described with reference to the flowchart of FIG. FIG. 4 is a flowchart showing the flow of pedestrian detection processing in the ECU of FIG. In the pedestrian detection apparatus 1, the operation described below is repeated at regular intervals.

  The visible light camera 10 images the front of the host vehicle with a long exposure time in accordance with the exposure control from the ECU 20, and transmits the captured image data to the ECU 20 as an image signal. The ECU 20 receives this image signal and acquires an image having a long exposure time (brightest) (S1). Further, the visible light camera 10 images the front of the host vehicle with an intermediate exposure time according to the exposure control from the ECU 20, and transmits the captured image data to the ECU 20 as an image signal. The ECU 20 receives this image signal and acquires an image having an intermediate exposure time (intermediate brightness) (S1). Further, the visible light camera 10 images the front of the host vehicle with a short exposure time according to the exposure control from the ECU 20, and transmits the captured image data to the ECU 20 as an image signal. The ECU 20 receives this image signal and acquires an image having a short exposure time (darkest) (S1).

  The ECU 20 selects an image having the longest exposure time (long exposure time) from the acquired images having different exposure times (S2). Then, the ECU 20 extracts a pedestrian candidate area that seems to be a pedestrian from the selected image (S3). Here, only one pedestrian candidate region or a plurality of pedestrian candidate regions may be extracted. If no pedestrian candidate area is extracted, the current process ends here.

  The ECU 20 selects one pedestrian candidate area from the pedestrian candidate areas extracted in S3 (S4). Then, the ECU 20 extracts a region having a luminance higher than the first luminance determination threshold as a high luminance region from the selected pedestrian candidate region, and calculates the center of gravity position of each extracted high luminance region (S5). Here, there may be a case where only one or a plurality of high luminance regions are extracted, and there is a case where none is extracted.

  The ECU 20 determines whether or not a high brightness area has been extracted in the process of S5 (S6). If it is determined in S6 that no high-intensity area has been extracted, the ECU 20 ends the process for this pedestrian candidate area and proceeds to S14.

  If it is determined in S6 that a high-luminance region has been extracted, the ECU 20 selects an image having the second longest exposure time (intermediate exposure time) from the images having different exposure times acquired in S1 ( S7). Then, the ECU 20 extracts a region having the same position and the same rectangular shape as the pedestrian candidate region selected in S4 from the selected image as a pedestrian candidate region with an intermediate exposure time, and within the extracted pedestrian candidate region. Then, a region having a luminance higher than the second luminance determination threshold is extracted as a high luminance region, and the barycentric position of each extracted high luminance region is calculated (S8). Further, the ECU 20 calculates the distance between the centroid position of the high brightness area in the long exposure time extracted in S5 and the centroid position of the high brightness area in the intermediate centroid position extracted in S8, and the distance is a distance determination threshold value. It is determined whether it is smaller than (S9). If it is determined in S9 that the distance between the centroid positions of the high luminance area is equal to or greater than the distance determination threshold, the ECU 20 determines that there is no high luminance area at the same centroid position, and the process for this pedestrian candidate area ends. The process proceeds to S14.

  If it is determined in S9 that the distance between the centroid positions of the high brightness area is smaller than the distance determination threshold, it is determined that there is a high brightness area at the same centroid position, and the ECU 20 determines that the images having different exposure times acquired in S1. An image having the third longest exposure time (short exposure time) is selected (S10). Then, the ECU 20 extracts a region having the same position and the same rectangular shape as the pedestrian candidate region selected in S4 from the selected image as a pedestrian candidate region with a short exposure time, and from within the extracted pedestrian candidate region. A region having a luminance higher than the third luminance determination threshold is extracted as a high luminance region, and the barycentric position of each extracted high luminance region is calculated (S11). Further, the ECU 20 calculates the distance between the centroid position of the high brightness area at the long exposure time extracted at S5 and the centroid position of the high brightness area at the short exposure time extracted at S11, and the distance is calculated from the distance determination threshold. It is also determined whether or not (S12). If it is determined in S12 that the distance between the center of gravity positions of the high brightness area is equal to or greater than the distance determination threshold, the ECU 20 determines that there is no high brightness area at the same center of gravity position, and the process for this pedestrian candidate area ends. The process proceeds to S14.

  If it is determined in S12 that the distance between the center of gravity positions of the high brightness area is smaller than the distance determination threshold, it is determined that there is a high brightness area at the same center of gravity position, and the ECU 20 is a pedestrian at three different exposure times. Since the barycentric positions of the high brightness areas in the candidate area are the same position, the pedestrian candidate area is determined as a non-pedestrian (light emitting object) (S13). Then, the ECU 20 determines whether or not the processing for all the pedestrian candidate areas extracted in S3 is completed (S14). When it determines with the process with respect to all the pedestrian candidate areas not having been complete | finished by determination of S14, ECU20 returns to the process of S4 and performs the process with respect to the next pedestrian candidate area | region.

  If it is determined in S14 that the processing for all pedestrian candidate areas has been completed, the ECU 20 determines that all pedestrian candidate areas extracted in S3 are not pedestrian candidate areas in S13. And the pedestrian candidate area is determined as a pedestrian (S15). And ECU20 calculates pedestrian information about the pedestrian candidate area | region determined to be a pedestrian, and transmits the pedestrian detection signal which consists of pedestrian information of each pedestrian to a driving assistance device.

  According to the pedestrian detection device 1, by comparing the characteristics of the shape of the high brightness area between the pedestrian candidate areas at the same image position in the images captured with three different exposure times (exposure conditions), the pedestrian A pedestrian can be detected with high accuracy even if there is a high-luminance region other than (light emitting object). In particular, the pedestrian detection device 1 can determine whether or not the feature of the shape of the high-brightness region has changed by comparing the position of the center of gravity of the high-brightness region between pedestrian candidate regions with three different exposure times. It can be judged with accuracy.

  Incidentally, since the pedestrian detection apparatus 1 uses the visible light camera 10 as an imaging means, it is also possible to discriminate a luminescent material that does not have a wavelength in the infrared light region. When an infrared camera is applied, there is a case where it is not possible to discriminate a light emitting object such as an LED that does not have a wavelength in the infrared region.

  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 a pedestrian detection device that detects a pedestrian, but it can also be applied to devices that detect other objects such as vehicles, bicycles, motorcycles, traffic signs, etc. in addition to pedestrians. is there. In this embodiment, the present invention is applied to a pedestrian detection device (object detection device) mounted on a vehicle. However, the present invention can also be applied to an object detection device other than on-vehicle such as mounting on a moving body such as a robot.

  Further, although a visible light camera is used in this embodiment, an infrared camera (near infrared camera or the like) may be used. In this embodiment, the front of the host vehicle is imaged with a visible light camera, and a pedestrian in front of the host vehicle is detected. However, with the visible light camera, other peripheral directions such as rear and side are imaged, The configuration that detects pedestrians in the peripheral direction is very good. In the present embodiment, each processing unit of the pedestrian detection device is configured by a dedicated ECU as an on-vehicle vehicle, but each processing unit may be configured as a general-purpose computer such as a personal computer.

  In the present embodiment, a configuration is used in which images are captured at three different exposure times, and the characteristics of the shape of the high-brightness region are compared between pedestrian candidate regions extracted from images of three different exposure times. However, two or four or more types of images with different exposure times are captured, and the characteristics of the shape of the high brightness region between the pedestrian candidate regions extracted from the images with two or more types of different exposure times, respectively. It is good also as a structure to compare. In the present embodiment, the exposure time is changed as the exposure condition, but other exposure conditions such as a lens aperture may be changed.

  In this embodiment, the centroid position distance between the high brightness areas respectively extracted from the pedestrian candidate areas with different exposure times is obtained and determined based on the distance. The number of high brightness areas respectively extracted from the candidate areas may be counted, and the determination may be made based on the number of high brightness areas in the pedestrian candidate area. In this case, it is determined whether or not the change in the number of high-intensity areas is equal to or less than a threshold value. If the change in the number of pedestrians is less than or equal to a threshold value, the pedestrian candidate area is determined as a luminescent object. As described above, in the case of a pedestrian, when the exposure time is shortened, the high-intensity area becomes smaller and is eventually divided, and the number of high-intensity areas changes. On the other hand, in the case of a luminescent material, even if the exposure time is shortened, the high luminance area is not so small and the number of high luminance areas does not change.

  DESCRIPTION OF SYMBOLS 1 ... Pedestrian detection apparatus, 10 ... Visible light camera, 20 ... ECU, 21 ... Pedestrian model database, 22 ... Exposure control part, 23 ... Multiple exposure image acquisition part, 24 ... 1st pedestrian determination part, 25 ... 1st DESCRIPTION OF SYMBOLS 1 High brightness area extraction part, 26 ... 2nd high brightness area extraction part, 27 ... 3rd high brightness area extraction part, 28 ... Luminescent substance determination part, 29 ... 2nd pedestrian determination part.

Claims (4)

An object detection device for detecting an object from an image,
Imaging means capable of imaging under different exposure conditions;
Comparison means for comparing the characteristics of the shape of the high-brightness region between object candidate regions at the same image position respectively extracted from a plurality of images with different exposure conditions imaged by the imaging means;
Determination means for determining whether or not the object candidate region is an object based on a comparison result in the comparison means;
An object detection apparatus comprising:   The determination means determines that the object candidate area is a luminescent object when a change in the feature of the shape of the high brightness area is small between a plurality of object candidate areas with different exposure conditions from the comparison result of the comparison means. The object detection apparatus according to claim 1.   3. The object detection apparatus according to claim 1, wherein the comparison unit compares the position of the center of gravity of the high brightness area between a plurality of object candidate areas having different exposure conditions.   4. The object detection apparatus according to claim 1, wherein the comparison unit compares the number of high-luminance areas between a plurality of object candidate areas having different exposure conditions. 5.

Images