JP4796860B2 - Object detection apparatus and object detection method - Google Patents

Description

  The present invention relates to an object detection apparatus and an object detection method for detecting an object to be inspected such as a surface defect from an input image.

  Conventionally, for example, in the inspection of a polarizing film or a retardation film, the surface of the object to be inspected is irradiated with light, and the reflected light is imaged by a line sensor. Techniques for detecting are known.

When a defect that is an inspection target object is detected using the one-dimensional waveform imaged by the line sensor, as shown in FIG. 6, whether the inspection target object is a noise or not with respect to the observed value V (x) at each position x. The threshold value _Vth is set as the first determination criterion for whether or not, and the area equal to or larger than the threshold value _Vth is extracted, the sizes (widths in this case) w1 and w2 are obtained, and the second determination criterion By comparing the threshold value _Wth set as and the area sizes w1 and w2, it is possible to determine whether the observed value V (x) is an inspection target object such as a coating defect or noise.

On the other hand, when an inspection object is detected using a two-dimensional image by an area sensor, areas (Area1 and Area2) exceeding the threshold _Vth for the luminance V (x, y) are extracted as shown in FIG. By setting the second threshold value _Sth for the area (S1, S2) of the region, it is possible to determine whether the object is an inspection object or noise. In addition to the area, this second threshold value may be set for a dimension of a circumscribed rectangle of the region, for example, wx1, wy1, wx2, wy2, etc.

In addition to the above-described noise identification method based on the size of the detection area, there is a method of tracking time changes. For example, as shown in FIG. 8, extracts the object a ₀ and the object b ₀ at a certain time t _0, to extract the object a ₁ and the object b ₁ which respectively move to the next time t _1, the following additional time t when extracting the object _{a 2} and the object _{b 2} and the object _{c 2} moved to _2, to predict the position of the object at time _{t 2} based on the object _{a 1} of _the object _{a 0} and time _{t 1} at time _{t 0} Thus, a ₂ closest to the predicted position can be determined as the object (see, for example, Patent Document 1). Further, an object that is not combined with any extracted object can be determined as noise. Alternatively, an object that forms a vector closest to a movement vector between time t ₀ and time t ₁ can be determined as the object at time t ₂ (among three vectors extending from a _1, from a ₀ vector closest to the vector directed to a ₁ is a vector from _{a 1} to _{a 2.).}

Japanese Patent Application Laid-Open No. 8-247794

  As described above, the method of identifying an object and noise based on the spatial size of an object extracted from a two-dimensional image is effective when the object and noise can be distinguished from each other. It is not effective when both are of the same size and only the temporal movement is different.

  Further, in the method for tracking the time change as described above, a high-level tracking algorithm is required, so that much processing time is required. In addition, the effect is great when a moving object is relatively easy to track, such as a person or a car, but when the random random noise (unnecessary extracted object) repeatedly occurs and disappears, tracking itself is For example, there may be a case where the correspondence cannot be determined between the past extracted object and the current extracted object.

  The present invention has been proposed in view of such a conventional situation, and provides an object detection apparatus and an object detection method capable of effectively detecting an inspection target object in an image with a small amount of calculation. With the goal.

  In order to solve the above-described problems, the present invention provides an object detection apparatus that continuously captures a two-dimensional image of an object to be inspected that moves in a predetermined direction and detects the object to be inspected from the two-dimensional image. Object candidate detection means for detecting the object candidate to be inspected for each two-dimensional image, and a candidate area including the object candidate in the two-dimensional image according to the object candidate detected by the object candidate detection means Candidate area generation means for generating and candidate area movement generation means for moving and generating the candidate area generated by the candidate area generation means according to the movement amount of the object to be inspected that moves in the predetermined direction for each of the two-dimensional images And the candidate area movement-generated by the candidate area movement generation means is the number of times of movement generation below the first threshold. If the number of times the candidate object is detected in the candidate area is greater than or equal to the second threshold among the number of movement generations of the candidate area, the object candidate detected in the candidate area is the object to be inspected. It is characterized by comprising determination means for determining that there is.

  Further, the present invention provides an object detection method for continuously capturing two-dimensional images of an object to be inspected moving in a predetermined direction and detecting the object to be inspected from the two-dimensional image. An object candidate detecting step for detecting the object candidate to be inspected, and a candidate region generating step for generating a candidate region including the object candidate in the two-dimensional image according to the object candidate detected in the object candidate detecting step; A candidate region movement generation step of generating and moving the candidate region generated in the candidate region generation step according to the amount of movement of the object to be inspected moving in the predetermined direction for each of the two-dimensional images; and the candidate region movement generation The candidate area generated by movement in the process has the number of movement generations equal to or less than the first threshold, and the candidate area is moved. A determination step of determining that an object candidate detected in the candidate area is an object to be inspected when the number of times the object candidate is detected in the candidate area is equal to or greater than a second threshold among the number of times of generation It is characterized by having.

  According to the present invention, an object candidate is detected for each two-dimensional image obtained by continuously imaging an inspection target object that moves in a predetermined direction, and a candidate area that includes the object candidate according to the detected object candidate Is generated in the two-dimensional image, the candidate area is generated for each two-dimensional image according to the movement amount of the object to be inspected, the candidate area is the number of times of movement generation equal to or less than the first threshold, and the candidate area If the number of times the object candidate is detected in the candidate area among the number of movement generations is equal to or greater than the second threshold, by determining that the object candidate detected in the candidate area is the object to be inspected, Even if there is not much difference in size between the object to be inspected and noise, the object can be detected effectively with a small amount of calculation without tracking the object. It is possible.

  Hereinafter, specific embodiments of the present invention will be described in detail with reference to the drawings. An object detection apparatus shown as a specific example of the present invention detects a defect location (or defect area) such as a scratch as an object from a sheet-like molded body such as a thermoplastic resin. Note that the object detection apparatus according to the present invention is not limited to the detection of defects such as scratches, and for example, it is also possible to detect a moving object from an image obtained by capturing a monitored landscape.

  The sheet-like molded body is formed by passing the thermoplastic resin extruded from the extruder through the gaps between the rolls to give smoothness and gloss to the surface, and taking the sheet while being cooled on the transport roll by the take-up roll. The Examples of the thermoplastic resin applicable to the present embodiment include methacrylic resin, methyl methacrylate-styrene copolymer (MS resin), polyolefins such as polyethylene (PE), polypropylene (PP), polycarbonate (PC), poly Examples thereof include vinyl chloride (PVC), polystyrene (PS), polyvinyl alcohol (PVA), and triacetyl cellulose resin (TAC). Examples of the thermoplastic resin sheet include single sheets and laminated sheets of these thermoplastic resins.

  In addition, examples of defects that are objects to be detected on the sheet-like molded article include so-called knicks caused by pressing flaws, bubbles at the time of molding, etc., so-called nicks caused by creases, etc., and so-called original fabric streaks caused by differences in thickness. Etc.

  FIG. 1 is a block diagram schematically showing the object detection apparatus 1. The object detection apparatus 1 is an area for capturing a two-dimensional image including a transporting unit 3 that transports a sheet-shaped molded body 2 that is continuous in the longitudinal direction with a constant width, and a reflection image or a transmission image by the illumination 4 of the sheet-shaped molded body 2. A sensor 5 and an analysis unit 6 that detects a defect that is an object to be detected from a two-dimensional image are provided.

  The conveying means 3 includes a sending roller and a receiving roller that convey the sheet-like molded body 2 in the conveying direction, and measures the conveying distance by a rotary encoder or the like. In this specific example, the conveyance speed is set to about 2 m to 12 m / min in the conveyance direction.

  The illumination 4 is arranged so that the width direction of the sheet-like molded body 2 is illuminated linearly, and a linear reflected image is reflected in an image photographed by the area sensor 5. The light source of the illumination 4 is not particularly limited as long as it emits light that does not affect the composition and properties of the single wafer film, such as a metal halide lamp, a halogen transmission light, and a fluorescent lamp. The illumination 4 may be provided at a position facing the area sensor 5 with the sheet-like molded body 2 interposed therebetween so that a transmission image appears in an image photographed by the area sensor 5.

  The area sensor 5 includes a CCD (Charge Coupled Device) or a CMOS (Complementary Metal-Oxide Semiconductor) that captures a two-dimensional image, and is arranged so as to capture a predetermined area of the sheet-like molded body 2 as shown in FIG. Has been.

  The analysis unit 6 includes an image input unit 61 that inputs the two-dimensional image data of the area sensor 5, a memory 62 that temporarily stores the input image data, and an image processing unit 63 that detects a defect based on the image data. These components are connected via a bus 64.

  Two-dimensional image data from the area sensor 5 input via the image input means 61 is buffered in the memory 62. The two-dimensional image data buffered in the memory 62 is subjected to defect detection processing by the image processing unit 63.

  As will be described later, the image processing unit 63 detects a defect candidate pixel that is an object candidate for each two-dimensional image captured by the area sensor 5, and a defect candidate that includes the candidate pixel according to the detected defect candidate pixel. A region is generated in the two-dimensional image. This defect candidate area is generated according to the amount of movement of the defect that moves in a predetermined direction by the transport means 3 for each two-dimensional image. Then, it is determined whether the object is a defect or noise based on the defect candidate pixel in the defect candidate area that is moved and generated according to the movement amount.

  Specifically, the defect candidate area that is moved and generated according to the movement amount has the number of movement generations that is equal to or less than the threshold Th1, and the defect candidate pixel is included in the defect candidate area out of the movement generation number of the defect candidate area. When the number of times of detection is equal to or greater than the threshold Th2 (Th2 ≦ Th1), it is determined that the defect candidate pixel detected in the defect candidate region is a defective pixel. Further, the number of movement generations of the defect candidate area that has been moved and generated exceeds the threshold Th1, and the number of movements of the defect candidate area that is detected in the defect candidate area is less than the threshold Th2. In this case, the defect candidate pixel detected in the candidate area is determined to be a noise pixel.

  Further, the above-described defect candidate pixel detection will be described. The processing steps are roughly divided into five processes: a threshold determination process, a binarization process, a labeling process, a feature extraction process, and a defect identification process.

  In the threshold value determination process, for example, a threshold value for the binarization process is determined using a discriminant analysis method based on similarity of the luminance histogram of the entire image or a p-tile method based on the area ratio of the object in the entire image. This threshold is preferably calculated for each frame. Thereby, robustness improves with respect to disturbances, such as the illumination intensity change. Note that a fixed threshold value may be used in a stable environment such as a room.

  In the binarization process, the two-dimensional image is converted into a binarized image based on the threshold value determined in the threshold value determination process. The binarized image can be obtained by separating the luminance histogram with the threshold value determined by the threshold value determination process.

  In the labeling process, the white area in the black background of the binarized image is labeled. Also, the black region in the white background of the binarized image is labeled.

  In the feature extraction process, the area of the white region and / or the black region, the ferret diameter, and the like are calculated. Also, those with a small calculated area and small ferret diameter are regarded as noise and excluded.

  In the defect identification process, a defect such as a point-like defect, a so-called knick, or an original fabric streak is identified based on the area or ferret diameter calculated in the feature extraction process. For example, a white area having an area larger than a specified value is excluded as an illumination image, a black area having an area larger than a specified value is excluded as a background, and the remaining area is identified as a defective area.

  Thus, by converting a two-dimensional image into a binarized image, a defective area can be detected with high accuracy.

  The object detection apparatus 1 having the above-described configuration transports the sheet-like molded body 2 in a certain direction by the transport unit 3, images the sheet surface illuminated by the illumination 4 in this transport process, and captures an image by the area sensor 5. Based on the obtained image data, the analysis unit 6 detects a defect which is an object to be inspected.

  Next, the defect detection process in the analysis unit 6 will be described with reference to the flowchart shown in FIG. This flowchart shows processing for each two-dimensional image input to the image processing unit 63.

  In step S11, the counter of the virtual frame in the two-dimensional image at a certain time t (frame) is incremented. As will be described later, this virtual frame is a defect candidate area that is generated for a defect candidate pixel that is detected as an object candidate to be detected, and disappears when the movement is generated a number of times that exceeds the threshold Th1.

  In step S12, the image processing unit 63 performs defect candidate pixel detection processing. This defect candidate pixel detection process detects a defect candidate pixel by performing the above-described threshold value determination process, binarization process, labeling process, feature extraction process, and defect identification process. The coordinates of the defect candidate pixels in the two-dimensional image at the time t are (x [t] [i], y [t] [i) (i = 0 to n), and the number is n [t] (n ≧ 0). ). In step S12. If a defect candidate pixel is detected, the process proceeds to step S13. If not detected, the process proceeds to the next defect process.

  In step S13, it is determined whether or not the defect candidate pixel detected in step S12 exists in the existing virtual frame. That is, it is determined whether or not the coordinates of the defect candidate pixel detected in the defect candidate area incremented in step S11 exist. When the defect candidate pixel exists in the virtual frame, the pixel counter of the virtual frame is incremented (step S15). If no defect candidate pixel exists in the virtual frame, a new virtual frame is generated (step S14), and the pixel counter is incremented (step S15). This virtual frame is generated, for example, as a region having a certain area around the defect candidate pixel.

  In step S <b> 16, the image processing unit 63 determines whether the pixel counter of the virtual frame is greater than or equal to the threshold value. If the pixel counter of the virtual frame is greater than or equal to the threshold Th2, the process proceeds to step S17 to determine whether or not the virtual frame counter is less than or equal to the threshold Th1. In step S16, if the pixel counter of the virtual frame is less than the threshold value Th2, the process proceeds to step S18 to determine whether the virtual frame counter is equal to or less than the threshold value Th1.

  If the virtual frame counter is equal to or smaller than the threshold Th1 in step S17, the image processing unit 63 determines that the defect candidate pixel detected in the virtual frame is a defective pixel (step S19). That is, the movement-generated virtual frame has the number of movement generations equal to or less than the threshold Th1, and the number of times the defect candidate pixel is detected in the virtual frame is equal to or more than the threshold Th2 (Th2 ≦ Th1). If it is, the defect candidate pixel detected in the virtual frame is determined to be a defective pixel. In step S17, when the virtual frame counter exceeds the threshold Th1, the virtual frame is extinguished and the defect candidate pixel detected in the virtual frame is determined to be a noise pixel.

  In step S18, when the virtual frame counter is equal to or smaller than the threshold Th1, the image processing unit 63 sets a defect candidate pixel detected in the virtual frame as a defect candidate, and proceeds to defect processing for the next two-dimensional image input. In step S18, when the virtual frame counter exceeds the threshold value Th1, the virtual frame is extinguished and the defect candidate pixel detected in the virtual frame is determined to be noise. That is, when the movement-generated virtual frame is the number of movement generations exceeding the threshold Th1, and the number of times the defect candidate pixel is detected in the virtual frame is less than the threshold Th2 among the movement generation times of the virtual frame, It is determined that the defect candidate pixel detected in the virtual frame is a noise pixel.

FIG. 3 is a schematic diagram for explaining a defect candidate pixel and a virtual frame. When the defect candidate pixel a is detected at time t ₀ (frame), the image processing unit 63 newly generates a virtual frame Fa (t ₀ ) centered on the defect candidate pixel a. The virtual frame Fa is generated for each time based on the conveying direction and the conveying speed of the conveying means 3, and after moving for a predetermined time, the virtual frame Fa disappears. FIG. 3 shows an example in which a virtual frame is generated for 9 frames and then disappears. Further, even when the defect candidate pixel b is detected at time t ₁ , the image processing unit 63 newly generates a virtual frame Fb (t ₁ ) centered on the defect candidate pixel b, and generates a movement between 8 frames. To do. Here, when the defect candidate pixel c is detected in the virtual frame Fa (t ₂ ) at time t ₂ , the pixel counter of the virtual frame Fa is incremented.

  In this way, for each two-dimensional image obtained by continuously capturing the detection target object moving in the predetermined direction, an object candidate is detected, and two candidate regions including the object candidate are determined according to the detected object candidate. The candidate area is generated according to the amount of movement of the object for each two-dimensional image, and the candidate area has the number of movement generations less than or equal to the threshold Th1. When the number of times the object candidate is detected in the area is equal to or greater than the threshold Th2, the size of the inspection object and noise is determined by determining that the object candidate detected in the candidate area is the object to be inspected. Even if there is not much difference, it is possible to detect the object effectively with a small amount of calculation without tracking the object.

  FIG. 4 shows the result of detecting point-like defects in the captured image of the surface of the sheet-like molded body 2. Here, each captured two-dimensional image has 512 pixels in the X-axis direction and 480 pixels in the Y-axis direction. In addition, when 50 frames are imaged at a speed of about 0.83 seconds and the lifetime of the virtual frame is 20 frames, and pixel candidates of 4 frames or more are extracted, object candidates (pixels) in the virtual frame are extracted. ) Is determined to be an object to be searched (defect). The sheet-like molded body 2 is conveyed in the Y-axis direction, and the Y coordinate value of the pixel representing the defect decreases with time. Except for pixels representing defects, there is no continuity and it is a single noise.

  Moreover, Fig.5 (a)-FIG.5 (e) are the examples of the two-dimensional image of the sheet-like molded object 2 which continued temporally. In the detection result object pixel group A shown in FIG. 4, there is a time (frame) at which no object candidate pixel is detected. This is a point-like shape on the surface of the sheet-like molded body 2 as shown in FIG. This is because the object candidate pixel was not detected due to irregular reflection of the defect or the like.

  As described above, according to the present invention, even when an object is not detected by the incident angle of illumination light, the object can be reliably identified from a plurality of temporally continuous two-dimensional images.

It is a functional block diagram which shows the structure of the principal part of the defect inspection apparatus to which this invention is applied. It is a flowchart which shows the object detection process for every two-dimensional image. It is a schematic diagram explaining a defect candidate pixel and a virtual frame. It is a figure which shows the result of having detected the point-like defect in the captured image of the surface of a sheet-like molded object. It is a figure which shows the example of the two-dimensional image of the sheet-like molded object which continued temporally. It is a schematic diagram explaining the detection of the object at the time of using the one-dimensional waveform imaged with the conventional line sensor. It is a schematic diagram explaining the detection of the object at the time of using the two-dimensional image by the conventional area sensor. It is a schematic diagram of the detection of an object when the time change of an object is tracked using the two-dimensional image by the conventional area sensor.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Object detection apparatus, 2 sheet-like molded object, 3 conveyance means, 4 illumination, 5 area sensor, 6 analysis part

Claims (7)

In an object detection apparatus that continuously captures a two-dimensional image of an object to be inspected that moves in a predetermined direction and detects the object to be inspected from the two-dimensional image,
Object candidate detection means for detecting the object candidate for inspection for each two-dimensional image;
Candidate area generating means for generating a candidate area including the object candidate in the two-dimensional image according to the object candidate detected by the object candidate detecting means;
Candidate area movement generation means for moving and generating the candidate area generated by the candidate area generation means according to the movement amount of the object to be inspected moving in the predetermined direction for each of the two-dimensional images;
The candidate area moved and generated by the candidate area movement generation means is the number of movement generations equal to or less than a first threshold, and the number of times the candidate object is detected in the candidate area is the number of movement generations of the candidate area. An object detection apparatus comprising: a determination unit that determines that an object candidate detected in the candidate region is an object to be inspected when the object value is equal to or greater than a second threshold value.   The determination means is the number of movement generations in which the candidate area moved and generated by the candidate area movement generation means exceeds the first threshold, and of the number of movement generations of the candidate area, the object is included in the candidate area. The object detection apparatus according to claim 1, wherein when the number of times a candidate is detected is less than the second threshold, the object candidate detected in the candidate area is determined to be noise. A transport means for moving the object to be inspected in a predetermined direction;
The object detection apparatus according to claim 1, wherein the candidate area movement generation unit moves and generates the candidate area according to a movement amount by the transport unit.   4. The object according to claim 1, wherein the object candidate detection unit detects the object candidate to be inspected based on a feature value obtained by binarizing the two-dimensional image. Detection device.   The object detection apparatus according to claim 1, wherein the object to be inspected is a defect generated in a sheet-like molded body. In an object detection method for continuously capturing two-dimensional images of an object to be inspected that moves in a predetermined direction and detecting the object to be inspected from the two-dimensional image,
An object candidate detection step for detecting the object candidate for inspection for each two-dimensional image;
A candidate region generation step of generating a candidate region including the object candidate in the two-dimensional image according to the object candidate detected in the object candidate detection step;
A candidate area movement generation step of moving and generating the candidate area generated in the candidate area generation step according to the movement amount of the object to be inspected moving in the predetermined direction for each of the two-dimensional images;
The candidate area moved and generated in the candidate area movement generation step is the number of movement generations equal to or less than a first threshold, and the number of times the candidate object is detected in the candidate area is the number of movement generations of the candidate area. And a determination step of determining that an object candidate detected in the candidate area is the object to be inspected when the value is equal to or greater than a second threshold value.   In the determination step, the candidate area moved and generated in the candidate area movement generation step is the number of movement generations exceeding the first threshold value, and the number of movement generations of the candidate area includes an object in the candidate area The object detection method according to claim 6, wherein if the number of times a candidate is detected is less than the second threshold, the object candidate detected in the candidate area is determined to be noise.

Images