FR2624608A1 - Device for automatic detection of defects in a mixed batch of objects - Google Patents

Abstract

The device for automatic detection of defects in a mixed batch of objects comprises a video camera 7 for observing the objects to be inspected 8, 8', means for fixing a luminance threshold corresponding to the observation of objects exhibiting defects, means for comparing the luminance signal delivered by the video camera when observing an object to be inspected with the said luminance threshold, and visual display or alarm means 19, selectively activated when the comparison means detect the exceeding of the said luminance threshold. The device furthermore comprises a cabled processor 30 for real-time and line-by-line processing of the video signal emitted by the camera, in order to produce a normalised video signal which is reset on each line to a set value which can be programmed as a function of the mean value of the luminance of the video signal for the line in question.

Description

Device for automatic detection of faults in a heterogeneous batch of objects.

 The object of the present invention is a device for automatically detecting faults in a heterogeneous batch of objects, comprising a video camera for observing the objects to be checked, means for fixing a luminance threshold corresponding to the observation of '' defective objects. means for comparing the luminance signal delivered by the video camera when observing an object to be checked and said luminance threshold, and display or alarm means selectively excited when the comparison means detect the crossing of said luminance threshold.

 The invention thus relates to the field of machine vision, using video cameras associated with processing or display systems, and more particularly machine vision techniques applied to quality control as well as sorting or classification. products such as pieces of wood for example.

 A device for automatic detection of faults by vision system implies a priori the processing of a very large number of pieces of information. Thus, for example, an image from a matrix camera of 512 points on 580 lines with 256 levels of gray represents more than 70 million elementary information to be processed in real time.

 To reduce the amount of information sent to the processing system so as not to saturate it, it is possible to transform the initial complex image into an image formed by dots defining only two levels (white level and black level ). This is achieved by comparing the luminance values of the various points of the image to a predetermined threshold. For each point, a white level or black level value is then assigned according to whether the luminance signal is higher or lower than the predetermined threshold.

 This type of pre-processing works well on homogeneous and perfectly illuminated images, but does not give satisfactory results as soon as there is in the field of the camera a heterogeneous batch of light products and dark products, since the threshold is identical all over the image.

 Insofar as the threshold allowing the detection of faults on light products is in fact different from that to be applied in the case of dark products, means can be implemented in order to create an adaptive threshold.

 According to a first possibility, the grayscale histogram is carried out and then the threshold is calculated. This solution is powerful since it makes it possible to carry out an operation of comparison with a threshold on a screen window, which makes it possible to calculate a threshold per object. The major drawback of such a solution lies in its execution time. Indeed, it is in this case necessary to proceed sequentially to an image acquisition, a definition of the windows, u-calculation of histogram for each window, a calculation of threshold for each window and an operation of comparison with the determined threshold for each window. These different operations imply a processing time which is not compatible with real-time processing.

 According to a second possibility, one proceeds by calculating the average gray level on the image using a wired processor.

Such a solution makes it possible to take account of the slow variations in the brightness of the scene, but does not offer any improvement as soon as light objects and dark objects are mixed in the same scene.

 According to yet a third possibility, the average luminance level is measured using a remote cell and connected to the vision system. This solution has the mena disadvantage that the previous one and, moreover, brings problems of interpretation because of its response time compared to the camera or because it integrates the background of the scene for example.

 The present invention aims to remedy the aforementioned drawbacks had to allow automatic and reliable real-time detection of faults in a batch of objects observed using a video camera, even if the batch of objects includes both light and dark objects.

 These aims are achieved by means of a detection device of the type defined at the head of the description, characterized in that it further comprises a wired processor for real-time and line-by-line processing of the video signal elected by the camera to produce a normalized video signal which is readjusted on each line to a programmable reference value as a function of the average value of the luminance of the video signal for the line in question.

 The video signal from a camera being representative of the light level reflected by the scene, the invention thus aims to process the video signal so as to obtain maximum homogeneity of the average luminance level, while retaining the original dynamics, which is possible due to the processing in real time and line by line. Obtaining a processed video signal thus normalized makes it possible to minimize both the problems due to variations in the hue of the samples in the mena scene, and the problems of homogeneity of lighting.

 The signal processor according to the invention allows the registration of the video signal, in real time, in order to authorize a comparison operation with a suitable threshold, not image after image, but line after line.

 This is done in a particular way by calculating the average value of the luminance of each line and locking it.

lage of this value on a programmable reference.

 According to a particular characteristic of the invention, in order not to involve the luminance values belonging to the background of the scene in the calculation of the average, two programmable limits are provided beyond which the calculation of the threshold is inhibited.

 This makes it possible not to involve in the calculation of the threshold the luminance values lower than the low limit, therefore belonging to the background in the case of a higher lighting (dark background), or the values higher than the high limit in the case diun lower lighting (bright background).

Other characteristics and advantages of the invention will emerge from the following description of a particular embodiment given by way of example, with reference to the appended drawings in which
FIG. 1 is a block diagram showing the main functions of an automatic fault detection device in a heterogeneous batch of objects, with a registration processor according to the invention,
FIG. 2 is a flowchart showing the succession of steps carried out in a processing processor according to the invention,
FIGS. 3 and 4 show the evolution of a luminance signal as a function of time, the shape of an average value and the positioning of a threshold, and
FIG. 5 is a block diagram of a registration processor usable in the context of a detection device according to the invention, such as that shown schematically in FIG. 1.

 FIG. 1 shows the overall diagram of a device for detecting faults in a heterogeneous batch of objects 8, which may consist, for example, of a mixture of light wooden boards and dark wooden boards. The objects 8 to be observed can for example be placed on a means of transport 80, such as a conveyor belt, which brings the batches of objects 8 successively into the field of a video camera 7 whose signals are applied to a processing system 9 intended to compare the luminance signals delivered by the video camera 7 during the observation of objects to be controlled 8, with a predetermined luminance threshold S fixed so as to correspond to the observation of objects exhibiting faults, such as the 8 'object for example.

Display or alarm means 19 connected to the processing system 9 are selectively excited when the comparison means detect the crossing of the luminance threshold S.

A detector 81 for the presence of objects in the field of the camera 7, and / or for controlling the movement of the conveyor 80 emits a signal
D which can be applied processing system 9 to allow the identification of the observed objects 8 'which have been the subject of a fault detection.

 As shown in FIG. 1, a preprocessing processor 30 receives the video signals emitted by the camera 7, which can for example be a black and white matrix camera of 512 points on 580 lines with 256 levels of gray. The pre-processed signals from the pre-processed cable processor 30 are then applied to the processing system 9 which may itself be conventional.

 The wired processor 30 essentially aims to carry out the real-time calculation of the average gray level line by line, and not image by image, in order to determine in real time a suitable threshold line by line which makes it possible to deliver to the processing system 7 a strongly pre-processed image.

 According to a first possible embodiment, the average value per line is weighted in order to constitute a suitable threshold allowing the obtaining of a binary image which is applied to the processing system 9 to be the subject of a comparison with the threshold predetermined S and allow the delivery of specific information in case of crossing of this threshold resulting in the presence of a defect in the field of vision of the camera.

 According to a second possible embodiment, the value of the average per line serves as a locking level (according to the so-called "clamping" technique) in a stage for realigning the infrared level of the video signal. The information produced by the wired processor and supplied to the processing system 9 is then a grayscale image, each line of which is weighted by the average value of the previous line.

FIGS. 3 and 4 represent an example of the shape of the luminance of a video signal V as a function of time over two successive lines for scanning a video image, FIG. 3 corresponding to a first line between the instants T1 and
TO and FIG. 4 corresponding to a second line included - between the instants TO and T + 1.

 In FIG. 3, S designates the threshold established so as to correspond to the average at time T-1 and M designates the average value of the video signal V determined by the processor 30 from the video signal V between the instants T-1 and TO.

 In FIG. 4, S designates the threshold established so as to correspond to the average at the time TO and M designates the average value of the video signal V determined by the processor 30 from the video signal V between the instants TO and T + 1.

 In Figures 3 and 4, L1 and L2 represent two low and high programmable limits beyond which the averaging is inhibited. This makes it possible not to involve in the calculation of the threshold the luminance values lower than the limit L1, therefore belonging to the background in the case of a higher lighting (black background) or the luminance values higher than the high limit L2 in the case of lower lighting (sky background).

 The main operations performed by the wired processor 30 are shown in the flow diagram of FIG. 2.

 During a first step 101, the processor measures the luminance of the input signal.

 In a second step 102, the processor performs a test to determine whether the measured luminance value is included or not within the predetermined limits L1 and L2 (FIGS. 3 and 4).

If this is not the case, there is a return to step 101. If on the other hand, the measured luminance value is indeed between L1 and L2, the processor proceeds to the next step 103 which includes the calculation of the average value of the luminance signal.

 After step 103, the processor performs the test 104 consisting in examining whether the end of a video line has been reached.

If this is not the case, there is a return to the initial step 101 of luminance measurement. If, on the other hand, the test 104 makes it possible to note that one has reached the end of a video line, there is passage to a step 105 of calculating the difference X constituted by the algebraic value of the difference between a setpoint and the average value last calculated in step 103. The next step 106 consisted in adding the algebraic value X to the output luminance signal in order to provide at the output of processor 30 a pre-processed signal readjusted on a setpoint, and therefore "normalized", which then makes it possible to overcome, during processing by the processing system 9, fluctuations due to problems of heterogeneity of lighting or variations in shade of the samples (by example planks or wooden sheets) of the same scene.

 FIG. 1 shows in the form of a block diagram the main functional modules of the processor 30 for registration of the luminance signal. The input signal from the video camera 7 is applied to a stage 1 of analog-digital conversion. The digital signals from stage 1 are applied on the one hand to a module 2 for calculating the average per video line, within low and high limits L1 and L2, and on the other hand, to a stage of correction 5. The algebraic value of the difference X between a set value provided by a memory 3 and the average value per video line coming from the calculation module 2 is determined by a subtractor module 4 and is applied to the stage of correction 5 to be added algebraically to the digital value from stage 1 of analog-digital conversion. The corrected signal coming from the correction stage 5 is applied to a digital-analog conversion stage 6 which provides a pre-processed analog output signal conforming to the standards of conventional video signals, to be applied to the processing system 9 which can itself be classic.

 FIG. 5 shows in a more complete way in the form of a block diagram the electronically realizable modules which ensure the elementary functions of the wired processor 30. The wired processor 30 for signal processing comprises first of all a separator circuit 10 receiving the signals delivered by the video camera 7 to provide on the one hand line and frame synchronization signals, on the other hand a luminance signal.

An analog-digital converter 11 receives the luminance signal from the separator circuit 10 as an input. A circuit 25 for storing the luminance values delivered by the analog-digital converter 11 is connected to sequencing logic circuits 12 controlled by said line and frame synchronization signals. A real-time averaging sub-assembly per video line comprises an adder circuit 26, a point counter 27 and a divider circuit 28. A subtractor circuit 4 for determining the value and the sign of the offset existing between the calculated average value in the sub-assembly 26, 27, 28, and delivered by the divider circuit 28 and a setpoint value contained in a setpoint memory 3 provides an offset value X to an adder circuit 5 which shifts the output signal of the analog converter -digital 11 by adding algebraically to the digital values from the converter 11 the offset value from the subtractor circuit 4.A digital-analog converter 6 transforms the numbers from the adder circuit 5 into an analog signal representing the luminance of the output signal, and an output amplifier 61 ensures the combination of the synchronization signals from the separator circuit 10 and the luminance signal from the conver digital-analog weaver 6 to recreate a CCIR composite video signal. The output amplifier 61 also makes it possible to adapt the impedance, for example 75 ohms.

 The wired signal processing processor 30 further comprises a high luminance limit value memory 21 and a low luminance limit value memory 22, as well as two comparators 23, 24. The first comparator 23 receives the signal on the one hand digital output from the analog-digital converter 11 and on the other hand the high limit value stored in the memory 21 while the second comparator 24 receives on the one hand the digital signal from the analog-digital converter II and on the other hand the value low limit stored in the memory 22. The outputs of the first and second comparators 23, 24 are connected to the luminance value memory 25 to limit the averaging to the luminance values between the high and low limits.

Claims (6)

1. Device for automatically detecting faults in a heterogeneous batch of objects, comprising a video camera (7) for observing the objects to be checked (8.8 ′), means for fixing a luminance threshold corresponding to l observation of objects presenting defects, means of comparison between the luminance signal delivered by the video camera during the observation of an object to be checked and said luminance threshold, and means of display or alarm ( 19) selectively excited when the comparison means detect the crossing of said luminance threshold, characterized in that it further comprises a wired processor (30) for real-time and line-by-line processing of the video signal emitted by the camera to produce a standardized video signal which is readjusted on each line to a programmable reference value as a function of the average value of the luminance of the video signal for the line in question. 2. Device according to claim 1, characterized in that the wired processor further comprises means for limiting the determination of the average value of the luminance of the video signal at each line to the luminance values comprised between a lower limit value (L1) and an upper limit value (L2). 3. Device according to claim 1 or claim 2, characterized in that the wired signal processing processor comprises a splitter circuit (10) receiving the signals delivered by the video camera to provide on the one hand synchronization signals line and frame, and on the other hand, a luminance signal; an analog-digital converter (11) receiving as input the luminance signal coming from the separator circuit (10); a circuit (25) for storing the luminance values delivered by the analog-digital converter (11), which circuit (25) for storing is connected to logic sequencing circuits (12) controlled by said synchronization signals line and frame; a real-time averaging sub-assembly per video line comprising an addition circuit (26), a point counter (27) and a divider circuit (28); a subtractor circuit (4) for determining the value and the sign of the offset existing between the calculated average value and a reference value contained in a reference memory (3), an adder circuit (5) for shifting the output signal of the converter analog-digital (11) by adding algebraically to the digital values from the converter (11) the offset value from the subtractor circuit (4), a digital analog converter (6) to transform the numbers from the adder circuit (5) into a signal analog representing the luminance of the output signal, and an output amplifier (61) ensuring the combination of the synchronization signals from the separator circuit (10) and the luminance signal from the digital-analog converter (6) to recreate a video signal CCIR composite. 4. Device according to claim 2 or claim 3, characterized in that the wired signal processing processor further comprises a memory (21) of high luminance limit value and a memory (22) of low luminance limit value, a first comparator (23) for receiving on the one hand the digital signal coming from the analog-digital converter (11) and on the other hand the high limit value stored in the memory (21), a second comparator (24) for receiving d on the one hand the digital signal coming from the analog-digital converter (11) and on the other hand the low limit value stored in the memory (22), the outputs of the first and second comparators (23, 24) being connected to the memory ( 25) luminance value to limit the averaging to luminance values between the high and low limits. 5. Device according to any one of claims 1 to 4, characterized in that it is applied to the observation of a heterogeneous batch of light and dark objects, such as boards or sheets of wood. 6. Device according to any one of claims 1 to 5, characterized in that the video camera is a matrix camera of at least 512 points on 580 lines.