JP5722266B2 - Light spot detector - Google Patents

Description

  The present invention relates to a light spot detection device that is an image processing device used as a sensor for performing light distribution control of an automobile headlight.

  Research and development of technology that automatically switches the high beam and low beam of the headlight by analyzing the video from the in-vehicle camera.

  Patent Document 1 describes a technology for detecting a taillight of a preceding vehicle or a headlight of an oncoming vehicle from a color camera image and switching the headlight from a high beam to a low beam so as not to dazzle the driver of the preceding vehicle or the oncoming vehicle. ing. The driver does not need to switch between the high beam and the low beam while paying attention to the preceding and oncoming vehicles, and can concentrate on driving. However, if the high beam and the low beam are switched instantaneously, the driver's field of view suddenly becomes brighter or darker, and if the control timing is not appropriate, the driver may feel uncomfortable.

  In Patent Document 2, this technique is further developed to calculate the distance from the headlight and taillight reflected on the camera to the preceding vehicle and the oncoming vehicle, and to continuously control the irradiation range of the headlight. It has been described. Compared with high beam and low beam two-stage control, more complex and appropriate control is possible. In addition, since the amount of light can be gradually changed, there is an effect of reducing the possibility of giving a sense of incongruity even if the control timing is somewhat inappropriate.

Japanese Patent Publication No. 6-55581 Special Table 2002-526317

  When a camera detects a vehicle light such as a headlight or taillight of an oncoming vehicle or a preceding vehicle with a camera, disturbance light such as a reflector, a traffic light, or a streetlight may be erroneously detected as the vehicle light. Since this directly leads to a sense of incongruity, how to reduce false detection is a technical issue.

  The range in which the glare occurs on the driver of the oncoming vehicle or the preceding vehicle with the high beam of the vehicle extends over 500m ahead, so in order to realize headlight control that does not cause glare, headlights of oncoming vehicles farther than 500m It is also necessary to detect the taillight of the preceding vehicle with a camera.

  However, distant headlights and taillights are difficult to detect because they are dark and small. This is because other noise light is detected only by increasing the exposure time. The taillight can suppress the erroneous detection of noise light by using the color information of red, but the headlight is easy to detect erroneously because the color information cannot be used. If binarization is simply performed at a predetermined threshold, the misdetected noise light is not only the street light in the vicinity, but also vending machines with large areas, light from stores, street lights, etc. are clearly reflected at the road surface. And light with unclear edges such as blooming of light from outside the angle of view.

  Conventionally, including the above-mentioned Patent Document 1 and Patent Document 2, these erroneous detections cannot be sufficiently dealt with.

  An object of the present invention is to provide a light spot detection device capable of detecting a headlight of an oncoming vehicle from a distance to a vicinity while suppressing erroneous detection under the condition where a night noise light source exists.

  The present invention separates the headlights of a distant oncoming vehicle and the headlights of a nearby oncoming vehicle and detects the light spot, and uses the headlights of the distant oncoming vehicle as a minute light spot, Light is detected as a saturated light spot. That is, according to the present invention, the headlight of a distant oncoming vehicle can remove noise light with a large area by subtracting a background image created by erasing minute light spots in an image (input image) taken by a camera from the input image. Noise light with an unclear edge is excluded as a background, and a minute light spot is detected as a light spot of a headlight of a distant oncoming vehicle. A headlight of a nearby oncoming vehicle detects a saturated area (saturated light spot) in an image captured by the camera as a light spot of a headlight of a nearby oncoming vehicle.

  More specifically, the present invention provides an image acquisition means for acquiring a first image obtained by photographing the periphery of the own vehicle, and a first image for generating a second image obtained by removing a minute light spot region from the first image. Image processing means, a second image processing means for generating a third image by subtracting the second image from the first image, and the third image at a first density threshold value. Third image processing means for generating a fourth image by binarizing, fourth image processing means for generating a fifth image by binarizing the light spot saturation portion from the first image, And fifth image processing means for obtaining a logical sum of the fourth image and the fifth image.

  According to the present invention, since the pre-processing is performed so as to leave the light of the headlight, the erroneous detection can be suppressed, and under the condition where the night noise light source exists, the detection of the oncoming vehicle from far to near while suppressing the erroneous detection. It becomes possible to detect the headlight.

It is a block diagram which produces | generates the binary image of the light spot of the light spot detection apparatus which concerns on one Example of this invention. It is explanatory drawing which produces | generates the binary image of the light spot of the light spot detection apparatus which concerns on one Example of this invention. It is a figure explaining the process of the minimum filter and the maximum filter in the Example of this invention.

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

  FIG. 1 is a block diagram for generating a binary image of a light spot in the light spot detection apparatus of the present embodiment. A minimum image 103 that expands the darkest part is applied to the input image 101 photographed by the camera, and then a maximum filter 104 that expands the brightest part is applied to generate a background image. The process of subtracting the background image from the input image 101 is performed by the difference filter 105, and the output of the difference filter 105 is binarized by the binarization filter 106 based on the first density threshold value, thereby allowing the light of a minute headlight to be obtained. Only points can be extracted.

  The process will be described with reference to a specific example of FIG. An image 201 is used as an example of the input image 101. In the image 201, it is assumed that the light 207 of the vending machine, the headlight 208 of the oncoming vehicle, and the light 209 of the streetlight inserted from outside the nearby angle of view.

  When this image 201 is binarized without performing pre-processing as in this embodiment, it becomes a binary image 204, which includes the light of a vending machine and the light of a streetlight that are noise lights. It will be binarized, and even if there is no headlight 208 for the oncoming vehicle, it will be erroneously detected as a headlight for the oncoming vehicle.

  In this embodiment, an image 202 is generated by applying the minimum filter 103 shown in FIG. The minimum filter 103 adjusts the size and frequency of the filter so that the headlight just disappears. Next, the Max filter 104 is executed to obtain an image 203. The maximum filter 104 has the same filter size and number of times as the minimum filter 103. Details of the minimum filter and the maximum filter will be described later. By setting the maximum filter 104 and the minimum filter 103 to the same size and number of times, the image 203 becomes a background image in which the headlight 208 has just disappeared from the image 201.

  Next, when a process of subtracting the image 203 from the image 201 by the difference filter 105 is executed, an image 205 in which only the headlight 208 remains is generated. When this image 205 is binarized with the first density threshold value, a binary image 206 is obtained. Here, the first density threshold is set to a value of about 100 for an 8-bit image, for example.

  In this way, by creating a background image that eliminates minute light spots in the input image captured by the camera, subtracting the background image from the input image, and binarizing the subtracted image, noise light with a large area and clear edges Noise light that is not used is excluded as a background, and only a minute light spot can be binarized. This makes it possible to detect headlights of distant oncoming vehicles while suppressing erroneous detection even under conditions where a night noise light source is present.

  With the above processing, a minute headlight, that is, a distant headlight, is extracted as a light spot without any problem, and light having a large area or light with unclear edges is removed. However, headlights of oncoming vehicles in the immediate vicinity have a large area or light spreads due to blooming and the edges are not clear, so they may be removed by the above processing. The headlights of an oncoming vehicle in the immediate vicinity are extremely bright and saturated, unlike the light of a vending machine or the light of a streetlight inserted from outside the angle of view. Therefore, in this embodiment, the saturated portion is supplemented by adding as a light spot. This makes it possible to detect headlights of oncoming vehicles from a distance to the vicinity while suppressing erroneous detection.

  In this embodiment, as shown in FIG. 1, the process of adding a saturated region as a light spot is performed by a saturated portion binarization filter 107 where the input image 101 is saturated, specifically, a density value. If it is an 8-bit image, it is performed by extracting pixels with sufficiently high brightness such as 240 or more. Finally, in the OR filter 108, a binary image obtained by ORing the binary image obtained by the binarization filter 106 and the binary image obtained by the saturation portion binarization filter 107 is used as the output image 102. Output.

  Here, the binarization threshold (second density threshold) of the saturated portion binarization filter 107 is compared with the binarization threshold (first density threshold) used in the binarization filter 106. By doing so, light spots with luminance lower than the binarization threshold used in the binarization filter 106 are forcibly excluded, and the binarization threshold used in the binarization filter 106 is exceeded. In addition, the light spot having a luminance lower than the binarization threshold value of the saturation unit binarization filter 107 is selected based on the area of the light spot, ambient light, or the like, and exceeds the binarization threshold value of the saturation unit binarization filter 107. Luminous light spots are forcibly extracted.

  By performing the processing shown in FIG. 1, a binary image is obtained in which a small light spot and a saturated light spot are set to 1 and the others are set to 0. Streetlights and reflectors are also small light spots, but these light spots are sufficiently darker than the headlights, and therefore can be eliminated by adjusting the first density threshold.

Here, the minimum filter 103 and the maximum filter 104 will be described with reference to FIG. For example, in the case of a typical 8-neighbor minimum filter of a 3 × 3 kernel, the pixel of I ₃₃ is calculated by Equation (1).

  In other words, the minimum luminance value is selected from a total of nine pixels including the calculated pixel and its eight adjacent pixels. By executing this process for all pixels, the dark part expands and the bright part is eroded. Here, the minimum filter 103 is executed not only once but repeatedly n times until the headlight 208 is extinguished. The number n is determined by the maximum radius r that the headlight 208 can take. For example, if r is 15 in equation (1), n = r = 15. Since r becomes larger as it gets closer, and further varies depending on the angle of view of the lens, the resolution of the image sensor, etc., n is determined after measuring the size of the headlight in the vicinity.

  There are other minimum filters, such as a minimum filter of 4 neighborhoods (calculated by (Equation 2)) and a minimum filter of 5x5 kernel (calculated by (Equation 3)).

  The 5x5 kernel has many reference pixels and costs more computation than the 3x3 kernel. However, since the number of iterations is half that of Equation (1), it can be processed at high speed depending on the hardware configuration. A wide range of 7x7 and 9x9 kernels may be used.

  For the maximum filter 104, the expression (4) is used when the minimum filter of the expression (1) is used, the expression (5) is used when the expression (2) is used, and the expression (6) is used when the expression (3) is used. Must be applied the same number of times.

  As is clear from the above description, the present invention can be summarized as follows. In other words, in order to eliminate noise light with a large area and noise with unclear edges, a background is created by erasing minute light spots, and a process of leaving only minute light spots by subtracting the background is performed before. As a process. By this process, light spots with unclear edges are excluded as background. However, the headlights of nearby oncoming vehicles may have a large area due to blooming and the edges may not be clear. Therefore, the saturated region is added as a light spot.

  In other words, the light spot detection device of the present invention includes an image acquisition unit that acquires a first image (input image 101) obtained by photographing the periphery of the vehicle, and a second light spot region that is obtained by removing a minute light spot region from the first image. First image processing means (minimum filter 103, max filter 104) for generating an image, and second image processing means (difference filter 105) for generating a third image by subtracting the second image from the first image ), Third image processing means (binarization filter 106) for binarizing the third image with the first density threshold value to generate a fourth image, and light spot saturation from the first image A fourth image processing means (saturated portion binarization filter 107) for binarizing a part with a second density threshold value higher than the first density threshold value to generate a fifth image; And a fifth image processing means (OR filter 108) that takes the logical sum of the image and the fifth image, That.

  According to the configuration of the present invention, it becomes possible to detect headlights of oncoming vehicles from far to near while suppressing false detection under the condition that night noise light source exists, and it is possible to control headlights more naturally. It becomes.

  In addition, this invention is not limited to the Example mentioned above, Various modifications are included. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. It is possible to add, delete, and replace other configurations for a part of the configuration of the embodiment.

  101 ... Input image, 102 ... Output image, 103 ... Minimum filter, 104 ... Max filter, 105 ... Differential filter, 106 ... Binary filter, 107 ... Saturation part Binary filter, 108 ・ ・ ・ OR filter

Claims (3)

Image acquisition means for acquiring a first image obtained by photographing the periphery of the vehicle;
First image processing means for generating a second image obtained by removing a minute light spot region from the first image;
A second image processing means for generating a third image by subtracting the second image from the first image;
Third image processing means for binarizing the third image with a first density threshold value to generate a fourth image;
Fourth image processing means for binarizing the light spot saturation portion from the first image by a second density threshold value higher than the first density threshold value to generate a fifth image;
A light spot detection device comprising: fifth image processing means for obtaining a logical sum of the fourth image and the fifth image. The light spot detection device according to claim 1,
The first image processing means is a light spot detection device having a minimum filter that expands a darkest part of an image and a max filter that expands a brightest part of the image. In the light spot detection apparatus of Claim 2,
The size and the number of times of the minimum filter and the size and the number of times of the maximum filter are equal to each other.

Images