DETECTION OF DEFECTS ON A HOMOGENEOUS SURFACE OF A MOVING OBJECT

Description

[0001] The invention relates to a method and a camera device for detecting defects on a homogeneous surface of an object during movement of the object according to the preamble of claim 1 and 15, respectively.

[0002] Conventionally, a CMOS image sensor is typically used for image acquisition. This sensor captures images at a specific frame rate, integrating all charges generated by incident photons within a given exposure time. However, with a moving object, the integrated photons of a pixel no longer originate from the same location on the object, depending on the exposure time, resulting in motion blur. Detecting defects presents a particular challenge: the frame rate must be fast enough to register a defect without it being lost in the motion blur. This, however, means that sufficiently long exposure times are no longer possible, especially at higher object speeds. The most suitable solution would be a line scan camera, which achieves a frame rate of several hundred kHz.However, a line scan camera has a very limited field of view, and the problem of extremely short exposure times remains.

[0003] Recently, a novel camera technology has emerged: the so-called event-based camera. It is also referred to as a neuromorphological camera, in analogy to the retina and visual cortex. In an event-based camera, there is neither a fixed frame rate nor a shared reading of pixels. Instead, each pixel independently checks whether it detects a change in intensity. Only in this case is image information generated and output, or read out, and only by that specific pixel. Each pixel thus acts as a kind of independent motion detector. A detected movement or other change in intensity is reported individually or asynchronously as an event. This allows the event-based camera to react extremely quickly to the dynamics of the scene. Images generated from these events are not as intuitive for the human eye to perceive because the static image components are missing.

[0004] An event-based camera is described, for example, in a white paper by the company Prophesee, which can be accessed on their website. Pixel circuits for an event-based camera are known from WO 2015 / 036592 A1, WO 2017 / 174579 A1, and WO 2018 / 073379 A1. EP 3 663 963 A1 describes the use of an event-based camera for code reading.

[0005] The current state of the art does not describe the use of an event-based camera for detecting defects on the homogeneous surface of a moving object. Simply replacing a conventional high-speed camera with an event-based camera is also insufficient. With an event-based camera, events could accumulate analogously to an exposure time. Then, as with conventional motion blur, the events would scatter across the field of view during the movement and could not be attributed to any specific defect. Conversely, the events could be processed into images on very short timescales. This emulates a conventional high-speed camera. Smaller defects would then merely generate isolated individual events that would be indistinguishable from noise.

[0006] European patent application EP 4 531 412 A1, published on April 2, 2025, discloses a camera device with an event-based image sensor for recording a stream of objects. The events are corrected according to the time elapsed since a reference point and the movement that occurred during that time; in other words, the motion is reversed based on the elapsed time in order to compensate for motion blur. However, the detection of disturbances is not provided.

[0007] US patent 11,138,742 B2 describes an event-based feature tracking method using optical flow determination. The approach is mathematically very complex, and again, no defect detection method is described.

[0008] Other relevant documents include: WO2024022571A1, published after the relevant priority date of the present application, which describes a system and method for detecting irregularities in an object; WO2023046510A1, which discloses a method for training a machine learning model and a programmable hardware unit; and the scientific publication by J. Xu et al. in IEEE Transactions on Instrumentation and Measurement, Vol. 69, No. 6, June 2020: "Surface Quality Assurance Method for Lithium-Ion Battery Electrode Using Concentration Compensation and Partiality Decision Rules" (XP011787721).

[0009] The object of the invention is therefore to improve the optical detection of defects on a homogeneous surface of a moving object.

[0010] This problem is solved by a method and a camera device for detecting defects on a homogeneous surface of an object according to claim 1 and 15, respectively. For this purpose, an image is captured, and it is immediately clear that, according to the invention, a data set of events is initially recorded, and an image is generated from this data set only later. The object is in motion relative to the recording position or the recording camera device. The object, which is inspected, for example, as part of a quality control check or in a manufacturing process, is preferably expected to have a completely homogeneous surface. Deviations from a homogeneous surface, which can have various causes such as microcracks or particles, are detected as defects. An evaluation reveals whether such defects are present and, preferably, at which locations.

[0011] The invention is based on the fundamental concept of capturing an image with an event-based image sensor and subtracting the intervening movement from the successively captured events. An event-based or neuromorphological image sensor has a multitude of pixel elements, for example, in a row or matrix arrangement. The differences to a conventional image sensor were briefly discussed in the introduction: The pixel elements detect changes in intensity instead of measuring the respective intensity as is usually the case. Such a change, which should preferably be sufficiently rapid and exceed a certain noise level, is one of the events that give the invention its name. Furthermore, signals are only provided or read out when such a change in intensity occurs, and only from the affected pixel or pixels.

[0012] The recording is made over a time interval long enough that the object's movement within this interval causes a potential disturbance to pass through several pixel elements. The pixel elements that detect a sufficient change in intensity trigger corresponding events at individually distinct times. These times are expressed, without loss of generality, as the time elapsed since a reference point. The reference point is, in principle, arbitrary, but preferably the beginning of the time interval. Choosing a different reference point is easily correctable; it merely results in a shift of the resulting image as a whole. During the elapsed time, the object has continued to move.

[0013] To compensate for this movement in the events, each event is calculated back to its location at the reference point. When an image is generated from these corrected events, it effectively approximates a stationary image rather than one captured in motion. However, since the actual image capture occurs in motion, the event-based pixel elements can respond to disturbances, whereas in a truly stationary state, no events would be generated at these disturbances, only pure noise.

[0014] The method is a computer-implemented procedure that runs, for example, in a computing unit of a camera and / or a computing unit connected to a camera.

[0015] The invention has the advantage that even the smallest changes caused by disturbances are detected through event-based recording. This event-based technology allows the inspection of rapidly moving objects with continuous data acquisition and highly accurate timestamps. Due to the motion compensation in the event evaluation according to the invention, even individual events or a small number of events are accumulated if they originate from a disturbance and therefore occur systematically and not just randomly due to noise, and thus can be detected in total.

[0016] The object in question is preferably a film used for battery production. The term "battery" is used broadly here to include rechargeable batteries or accumulators. Such films fulfill the requirement of a homogeneous surface under undisturbed conditions. However, undetected defects pose a significant problem for the manufactured battery; inadequate film performance leads to heat and capacity losses, and in extreme cases, even fires and explosions.

[0017] The object is preferably an anode foil, a cathode foil, or a separator foil. This allows all foils relevant to battery manufacturing to be inspected for defects. Advantageously, several or all types of foils are inspected for defects during production, particularly directly within the manufacturing process, thus preventing defects caused by these defects from occurring in the first place.

[0018] Preferably, particles with a diameter of at most half the film thickness are detected as defects, especially particles with a diameter of no more than half the film thickness. Larger defects are thus also detected; the challenge lies in detecting the smallest structures. The order of magnitude, as usual, means an allowable factor n or 1 / n with n < 10 relative to the film thickness, i.e., for example, double or triple the thickness. Half the thickness is particularly significant because even smaller particles cannot penetrate a film, and therefore such defects are not necessarily detected. In absolute terms, for typical films, especially battery films, this means a size on the order of a few micrometers, for example, 5 µm, 10 µm, 15 µm, or 20 µm.

[0019] Preferably, both the top and bottom surfaces of the film are captured and inspected for defects. Defects are thus detected on both sides. For this purpose, preferably two image sensors or cameras are used, although mirror constructions in conjunction with a split field of view of a single image sensor are also conceivable.

[0020] The object is preferably illuminated differently in several lighting zones for image capture. For this purpose, several lighting units or several modules of a lighting unit are preferably provided. The field of view of the image sensor is thus divided into the different lighting zones. This allows defects to be detected in different lighting scenarios and therefore with an even lower error rate. The lighting zones are preferably non-overlapping with each other; otherwise, additional lighting zones are effectively created in the overlapping areas due to the superposition of light. Depending on the embodiment, overlapping areas can be taken into account as such in the evaluation or excluded. The advantage of the lighting zones is that they can be generated simultaneously.With complete illumination, the lighting could only be varied sequentially, which would result in recording at different stages of movement. However, it is not impossible that lighting zones, units, or modules could be activated in a temporal pattern.

[0021] Preferably, corrected events are collected only within the same illumination zone. This is easily implemented because the pixel elements of the image sensor are directed at a specific location and can therefore be assigned to an illumination zone. Without this assignment, the different illumination scenarios would be mixed together. Even with such averaging effects, dividing the image into illumination zones can be helpful, but the best results are achieved when the illumination scenarios are evaluated individually. The ability to activate illumination zones simultaneously makes this possible without any loss of time.

[0022] The illumination zones preferably differ from one another in at least one of the following illumination properties: incident or transmitted light, polarization, intensity, spectrum, angle of incidence, or illumination pattern. This can be achieved through appropriate properties of illumination units or modules. Detection in incident light is preferably performed from both sides. While rarely necessary for transmitted light, this is conceivable because a defect on either side of a semi-transparent object can produce a different effect. By varying the other aforementioned illumination properties, defects of various types are made visible, thus improving the overall detection rate.

[0023] The time interval preferably corresponds to the duration within which the homogeneous surface moves through an illumination zone or the entire field of view of the image sensor. This allows the maximum number of events to be captured, namely those in the same illumination scenario or in total. This also improves the detection rate for defects.

[0024] The movement occurring during the time interval is preferably predefined by a parameter setting, measured using an additional sensor, or determined from the events. Parameterization allows for the compensation of a fixed, predefined movement, for example, by specifying a speed for uniform motion, and more complex movements can also be parameterized. It is conceivable to adopt the movement parameters from the settings of a conveyor system that moves the object being picked up or, for example, unwinds or winds a film. Measurement with an additional sensor, such as an incremental encoder on a conveyor system, requires slightly more effort but is more flexible and adaptable. A third option is to determine the movement from the image sensor's measurement data, namely the events.In this case, there are no specifications or additional information about the movement, which is instead estimated independently. This eliminates the need for interfaces, and, as with measurements using an additional sensor, ensures that the actual movement, and not merely a desired or specified one, is taken into account.

[0025] Preferably, the object moves in only one direction. This is a common application and significantly simplifies event correction. The movement is particularly uniform, meaning it occurs at a constant speed. The application example of a conveyor system or the winding and unwinding of films has already been mentioned; here, the movement is generally uniform except for occasional switching operations. The speed is a few meters per second, for example, in the range of 0.5–1.5 m / s, corresponding to a typical conveying speed or running speed of a film.

[0026] The image sensor is preferably aligned with the rows or columns of pixel elements in the direction of movement. The camera is therefore positioned relative to the object stream so that the movement follows the rows or columns of pixel elements on the image sensor. This facilitates further processing and avoids discretization artifacts.

[0027] The events are prioritized according to the regulation. X n = X - v * dT corrected, with X Position of the pixel element that triggered the event, X n new position, v speed of movement and dT Time elapsed since the reference point. The direction of motion is generally referred to as the X-direction; this can always be achieved by a simple rotation if the direction of motion does not coincide with the lines of the image sensor. No correction is necessary in the Y-direction because the object stream does not move in this direction. The units are selectable; for example, the positions correspond to the pixel positions on the image sensor, the elapsed time is given in seconds, and the speed is given in pixels per second.

[0028] For correction based on the time elapsed since a reference point and the movement that occurred during that time, the events themselves are preferably evaluated without initially assembling an image. Only after the correction are the corrected events collected in an image, which is then analyzed for disturbances. Events occur only rarely when recording a homogeneous area. Therefore, it is advantageous to keep the analysis at the event level for as long as possible (sparse data, sparse evaluation). It is even conceivable to find the disturbances at this event level as well, for example, as clusters of events, and thus never assemble an image. However, this precludes the use of established image analysis methods.

[0029] The image is preferably evaluated using a machine learning method, in particular a neural network, to determine whether it contains defects. The greater challenge in detecting small defects is obtaining sufficient visual information. Once an image is available, thanks to the invention, in which the defects are visible, methods suitable for conventionally captured images can be used for the final defect detection. This is particularly flexible with machine learning, preferably employing a neural network or, even more preferably, a deep convolutional network. Supervised learning with annotated or labeled training data is particularly suitable for training. For this purpose, objects of the type to be examined, especially transparencies, are presented, and an external evaluation is provided for each object, indicating whether and, if so, where defects are present.The machine learning process can then generalize during later operation.

[0030] Traditional image analysis is also possible. The image is expected to be largely empty, at least after applying a noise threshold, since a homogeneous area without defects will not trigger any events. Therefore, one can search for clusters of events in which a minimum number of events were detected at a pixel of the resulting image, and / or with the requirement that events are detected at a minimum number of adjacent pixels in the image. It is important to note that in this context, "pixels" refers to the image points in the image generated from the corrected events; these should not be confused with the pixel elements that triggered the events.

[0031] An event preferably contains coordinate information of the associated pixel element, time information, and / or intensity information. A conventional image sensor data stream consists of the intensity or grayscale values ​​of the pixels, and the spatial reference in the image sensor plane is established by reading all pixels in an ordered sequence. In contrast, the event-based image sensor preferably outputs data tuples for each event, making the event assignable. These preferably include the location of the associated pixel element, such as its XY position on the image sensor, the polarity or direction ±1 of the intensity change, or the intensity measured at the time of the event, and / or a timestamp. As a result, very little data needs to be read out despite the high effective frame rate.

[0032] The event-based image sensor generates image information preferably at a refresh rate of at least 1 kHz, or even at least 10 kHz or more. The refresh rate of a conventional camera is the frame rate. An event-based camera does not have such a common frame rate, as the pixel elements output or refresh their image information individually and based on events. This allows for extremely short response times, which would only be achievable with a conventional camera at immense cost with a thousand or more frames per second. Even with an event-based sensor, an even higher refresh rate would be technically impossible to achieve with conventional cameras.

[0033] Each pixel element preferentially detects when the intensity it detects changes, and only then does it generate an event. This, in other words, expresses the special behavior of the pixel elements in an event-based camera or image sensor, which has already been discussed several times. The pixel element checks whether the detected intensity is changing. Only this is an event, and only an event triggers the output or reading of image information. A kind of hysteresis is conceivable, in which the pixel element ignores a defined, insignificant change in intensity and does not interpret it as an event.

[0034] The pixel element preferably provides differential information as image data, indicating whether the intensity has decreased or increased. Image information read from the pixel element is, for example, a polarity, a sign +1 or -1 depending on the direction of the intensity change. A threshold for intensity changes can be set, below which the pixel element does not trigger an event. The duration of an intensity change can also play a role, for example, by adjusting the reference value for the threshold with a certain decay time. A change that is too slow will then not trigger an event, even if, over a longer time window than the decay time, the total intensity change exceeded the threshold.

[0035] As an alternative to a differential event-based image sensor, an integrating variant is also conceivable. In this case, the pixel element provides integrated intensity as image information within a time window determined by a change in intensity. Here, the information is not limited to one direction of intensity change; instead, the incident light is integrated within a time window defined by the event, thereby determining a gray value. The measured value thus corresponds to that of a conventional camera, but the time of acquisition remains event-based and linked to a change in intensity.

[0036] The camera device according to the invention for detecting defects on a homogeneous surface of an object, in particular a film, during movement of the object relative to the camera device, has an event-based image sensor and is therefore an event-based or neuromorphological camera and consequently not a conventional camera with a conventional image sensor. A control and evaluation unit of the camera device executes one of the embodiments of the method according to the invention. The control and evaluation unit can be integrated into the camera, external, or a hybrid of both.

[0037] The invention is further explained below with regard to additional features and advantages by way of example embodiments and with reference to the accompanying drawing. The illustrations in the drawing show: Fig. 1 a camera in an application over a moving film; Fig. 2 a schematic representation of a time-dependent intensity profile to illustrate the operating principle of an event-based camera; Fig. 3 an exemplary position-time diagram of events; and Fig. 4 a schematic representation to illustrate lighting with multiple lighting zones.

[0038] Figur 1 Figure 10 shows a camera 10 that records an object 14 with a homogeneous surface 16 moving in one direction 12 within its field of view 18. In this example, the object 14 is a film, specifically a battery film, which is moved or rolled up or down by means not shown. Alternatively, objects 14 on a conveyor belt or other moving objects 14 are conceivable. By using additional cameras and / or mirrors and the like, wider objects 14 or other surfaces, such as the underside of the object 14, can also be recorded. The inspection of the object 14 with the camera 10 aims to detect, or preferably locate, possible defects 17 in the homogeneous surface 16. Figur 1 The fault 17 shown is greatly exaggerated in size for illustrative purposes.

[0039] The camera 10 uses an image sensor 20 and a lens 22 of any known design (shown only schematically) to capture image information of the moving object 14. The image sensor 20 typically comprises a matrix or row arrangement of pixels and is an event-based image sensor. Unlike a conventional image sensor, where charges are collected across a certain integration window and the pixels are then read out together as an image, the individual pixels trigger and transmit events when a change in intensity occurs within their field of view. The principle of an event-based image sensor will be explained later with reference to the Figur 2 explained in more detail.

[0040] A control and evaluation unit 24 is connected to the image sensor 20. This unit controls the sensor's recordings, reads out the respective events, and processes them further. The control and evaluation unit 24 comprises at least one digital processing component, such as at least one microprocessor, at least one FPGA (Field Programmable Gate Array), at least one DSP (Digital Signal Processor), at least one ASIC (Application-Specific Integrated Circuit), at least one VPU (Video Processing Unit), or at least one neural processor. Furthermore, the control and evaluation unit 24 can be located at least partially external to the camera 10, for example, in a higher-level control system, a connected network, an edge device, or a cloud.

[0041] The camera 10 outputs information via an interface 26, such as image data or evaluation results derived therefrom, in particular the presence or absence of a defect 17, preferably including its position. If the functionality of the control and evaluation unit 24 is at least partially located outside the camera 10, the interface 26 can be used for the necessary communication. Conversely, the camera 10 can receive information from other sensors or a higher-level control system via the interface 26 or another interface. This makes it possible, for example, to transmit a fixed or current velocity of the object 14's movement (measured by an additional sensor not shown) to the camera 10, or to obtain geometric information about the objects 14, in particular their distance from the camera 10 for focusing or other adjustments of the lens 22.The camera 10 can contain additional elements or be connected to additional components, for example to a sensor such as a light barrier or a light switch, which triggers the start or end of recording.

[0042] If the object 14 to be inspected is a non-transparent film, it is preferably inspected from both sides to detect defects 17 on the underside as well. A transparent film can be inspected using a transmitted light method. Specifically in the case of battery production, there are three types of films that can be inspected: the cathode film, the anode film, and the separator film. To find all defects 17, preferably six homogeneous surfaces 16 are inspected, namely the top and bottom surfaces of all three types of films. This is preferably done early in the production process so that the finished batteries only have films without defects 17. A typical size of the defects 17 to be detected is 5 µm, corresponding to half the thickness of a separator film, at movement speeds of 0.5–1.5 m / s.If several cameras 10 are required to cover a wider object 14 or for additional perspectives, for example from below, the functionality of the control and evaluation unit 24 can be distributed almost arbitrarily. It should also be mentioned that cut edges can be inspected in the same way.

[0043] Figur 2 To illustrate the operating principle of the event-based image sensor 20, the upper part shows a purely exemplary temporal intensity profile in a pixel element of the image sensor 20. A conventional image sensor would integrate this intensity profile over a predefined exposure time window, the integrated values ​​of all pixel elements would be output at a predefined frame rate and then reset for the next image.

[0044] The pixel element of the event-based image sensor 20, on the other hand, reacts individually and independently of a frame rate to a change in intensity. Vertical lines mark the times at which a change in intensity was detected. In the lower part of the Figur 2 At these times, events are represented with plus and minus signs depending on the direction of the intensity change. It's conceivable that the pixel element doesn't react to every intensity change, but only when a certain threshold is exceeded. Furthermore, it might be required that the threshold be exceeded within a specific time window. Comparison values ​​of previous intensities outside this time window are then effectively ignored. Advantageously, a pixel element can be individually configured, at least roughly, for lightning detection by defining a threshold and / or time window.

[0045] The events generated by a pixel element are read out individually at the time of the event or, preferably, in readout cycles of duration dt and thus transmitted to the control and evaluation unit 24. The time resolution is by no means limited to dt, as the pixel element can assign a timestamp of any desired precision to each event. The cycles defined by dt are also not comparable to a conventional frame rate. Conventionally, a higher frame rate means a directly linear increase in the amount of data due to the additional images. With the event-based image sensor 20, the amount of data to be transmitted does not depend on dt, except for a certain administrative overhead. If dt is chosen to be shorter, fewer events need to be processed per readout cycle. The amount of data is determined by the number of events and is therefore largely independent of dt.

[0046] In addition to differential event-based cameras, there are also integrating event-based cameras. They react to changes in intensity in a completely analogous way. However, instead of outputting the direction of the intensity change, they integrate the incident light within a time window defined by the event. This results in a grayscale value. Differential and integrating event-based cameras have different hardware architectures, and the differential event-based camera is faster because it does not require an integration time window. For further information on event-based camera technology, please refer again to the patent literature and scientific literature mentioned in the introduction.

[0047] The image information from the event-based image sensor 20 is not an image, but rather an event list. Each event is output, for example, as a tuple containing the sign of the intensity change in the case of a differential event-based camera or a gray value in the case of an integrating event-based camera, the pixel position on the image sensor 20 in the X and Y directions, and a timestamp. The correction of events for temporary movement, as described below, can initially be performed at the event or event list level. An image in the conventional sense for fault detection is preferably generated only at the end from the already corrected events.

[0048] Figur 3 Figure 1 shows an exemplary position-time diagram of events from a few selected pixel elements of the image sensor 20 that are adjacent to each other in the direction 12 of the movement. The X-axis represents the position of the pixel triggering each event, where the X-direction, without loss of generality, is the row direction and the direction of movement. The Y-axis is the time axis. Each point in the position-time diagram thus corresponds to an event in a pixel at position X at the trigger time t. The position-time diagram shown serves for illustration purposes; the control and evaluation unit 24 can also work directly with event lists or in any other representation.

[0049] A fault point 17 triggers events in neighboring pixels one after the other as it moves. Figur 3 The disturbance 17 is so small that only one pixel triggers at a time. For uniform motion, a straight line appears in the position-time diagram, highlighted by brighter dots for clarity. The slope of this line is proportional to the speed of the uniform motion, or, in the chosen units of the axes, equal to the speed. The motion speed can be estimated from this line; it can be either predefined or measured by an additional sensor. To ensure that the slope of the line or other speed information is usable for subsequent steps, calibration should be performed. For a lens 22 with strong geometric distortion, nonlinear calibration using a polynomial fit or similar method may be necessary.To account for different speeds, multiple calibrations or, in the non-linear case, interpolations can be performed.

[0050] Knowing the velocity, for example in pixels per second, now makes it possible to correct the movement of the disturbance 17 in the recorded events. Figuratively speaking, the events are to be converted to a common reference point based on the known intermediate movement. In the following, without loss of generality, the reference point is the beginning of a time interval in which events are collected for an image capture. This choice is ultimately arbitrary and not particularly significant, since a different reference point merely produces a common offset of the entire motion-compensated image.

[0051] For each event, the previous value X of the event can then be converted into a corrected value. X n be converted according to the regulation X n = X - v * dT. This is v the speed, which is determined in particular as a gradient, preferably in the unit pixels / s and dT The time elapsed since the reference time, according to the trigger time of the event under consideration. No correction is necessary in the Y-direction, as the movement occurs in the X-direction. After this conversion, the trigger time of the event is no longer of interest, at least for the method according to the invention, but would of course still be available for other evaluations.

[0052] In a slightly more complete notation, consider an event e at time t at position (x, y) of the image sensor 20, which has polarity p. The event is transformed according to the above calculation rule according to the velocity v: e ( t, x n , y, p ) = e ( t, x - v * dT, y, p All events are subjected to this transformation. An FPGA is particularly well-suited for at least some of the necessary calculations, which are frequently performed, not particularly complex in themselves, and easily parallelized. The new x-coordinate is usually not an integer and can then be rounded. From an implementation perspective, especially in an FPGA, it can be advantageous not to calculate with floating-point numbers, but to choose a format with fixed decimal places and to implement rounding and the like using bit shift operations.

[0053] The events corrected in this way can now be collected in a matrix based on their (x,y) coordinates; the result then corresponds to a conventional image. Thanks to the correction for motion, events triggered by a disturbance 17 contribute to the pixels of this image at the same locations. This makes even small or poorly visible disturbances 17 detectable. Such a disturbance 17 does not necessarily trigger an event in every pixel element of the image sensor 20. However, as the motion progresses, the disturbance 17 successively enters the field of view 18 of, for example, 720 pixel elements of an image sensor row, assuming, without loss of generality, that the motion is oriented in the line direction of the image sensor 20.Even a poorly discernible anomaly 17, which, for example, triggers an event only every tenth time in a pixel element, accumulates into a total signal of 72, and this is clearly distinguishable from noise. The motion correction ensures that this total signal becomes cumulative, unlike the previously very sparsely distributed individual events in every tenth pixel of the computational example, which were still below the noise threshold.

[0054] In the example just explained, it was implicitly assumed that each event contributes a polarity of +1, meaning the events are simply counted. This results in a grayscale image where the brightness corresponds to the number of events. Such a grayscale image can be analyzed, for example, using a machine learning method, particularly a neural network, to determine whether a defect 17 is present. Alternatively, conventional image analysis methods can be used. After correction, as explained, the grayscale value generated by a defect 17 is sufficiently distinct to reliably differentiate it from the otherwise homogeneous background, which is characterized only by noise.

[0055] There are alternative ways to collect the corrected events in an image. For example, events can be summed taking their polarity into account, so that positive and negative events cancel each other out. It is also possible to subtract an absolute number from a sum calculated taking the sign into account, revealing where in the image a particularly large number of events have canceled each other out. This might indicate an unstable or unreliable image feature. Furthermore, an image consisting only of positive events and / or only of negative events is conceivable. Polarity thus allows for the selective highlighting of different aspects in the image that may be important for subsequent fault detection.

[0056] The time interval in which events are collected for a given image is essentially a free parameter. A preferred upper limit is the time an object needs to traverse the field of view. However, fractions of this time are also conceivable, for example, to focus on specific sub-areas or objects, or to generate two or more consecutive images for each traverse time of an object. Such sub-areas could, for example, be different lighting zones, which are discussed further below in connection with Figur 4 This will be explained in more detail later. In the direction of movement, the image sensor 20 has a total of N pixels for the entire field of view 18 or a corresponding fraction for a partial area. The speed of movement in pixels / second, already discussed above, simply needs to be multiplied by N to determine how long events are collected for an image. Preferably, the evaluation is performed on a rolling basis, meaning a new image is generated and evaluated each time the object 14 has moved one or more pixels, with the oldest events being discarded accordingly. The dynamic range of the image, i.e., the possible gray values, corresponds to N, since each edge of a defect 17 triggers an event at most once in each pixel element.

[0057] Figur 4Figure 1 shows a schematic representation illustrating illumination with multiple illumination units or modules 28a-b, where only two modules 28a-b are shown as examples. The multiple modules 28a-b enable the creation of different illumination zones 30a-b. Preferably, events in the image to be evaluated are collected within an illumination zone 30a-b. Alternatively, it is conceivable to collect events across the boundaries of illumination zones 30a-b, which leads to an averaging effect. By varying the arrangement and properties of the modules 28a-b, a wide variety of illumination scenarios can be implemented. These scenarios differ, for example, spectrally (i.e., in color), polarization, angle of incidence, profile, or illumination pattern (such as lines, points, grids, homogeneous, and many more), as well as in intensity or its temporal behavior (such as modulation or gradients).Thus, darkfield, transmitted light, incident light, and other techniques can be used to visualize a variety of different defects 17. Polarization is understood to be relative to the receiving path; that is, preferably, in the case of polarized illumination, a polarizing filter is also associated with the image sensor 20. In particular, detection can be performed under crossed polarization, i.e., with illumination whose polarization direction differs from that of a polarizing filter in the receiving path, even more preferably perpendicular to each other. The multiple illumination scenarios significantly increase the probability that a defect 17 can actually be detected. The different illumination scenarios can be implemented simultaneously. This is a major advantage over a conventional line scan camera, which could only vary illuminations sequentially, thus further increasing the requirements for its acquisition frequency.As already discussed in the introduction, only a line scan camera is even an option, since a conventional matrix camera has a much too low recording frequency.

Claims

1. A method for recognizing defects (17) on a homogeneous surface (16) of an object (14), in particular a film, during a movement of the object (14), wherein an image of the homogeneous surface (16) is recorded and the image is evaluated as to whether it has defects (17), characterized in that the image is recorded by an event-based image sensor (20) which detects events of a changing intensity with a plurality of pixel elements; in that the recording takes place over a time interval within which the homogeneous surface (16) moves on over at least some pixel elements of the image sensor (20); in that the events are corrected in accordance with a time that has elapsed since a reference point in time and the movement that has taken place during the elapsed time; and in that the corrected events are collected in an image which is then evaluated with respect to the defects (17).

2. A method in accordance with claim 1, wherein the object (14) is a film for manufacturing a battery.

3. A method in accordance with claim 2, wherein the object (14) is an anode film, a cathode film, or a separator film, wherein in particular the anode film, cathode film, and / or separator film are checked for defects (17) during the manufacture of a battery.

4. A method in accordance with any one of the preceding claims, wherein particles in the order of magnitude of the thickness of a film, in particular particles from a diameter onward of at most half the thickness of the film, are already recognized as defects (17).

5. A method in accordance with any one of the preceding claims, wherein the lower side and the upper side of the film are recorded and checked for defects (17).

6. A method in accordance with any one of the preceding claims, wherein the object (14) is differently illuminated for the recording of the image in a plurality of illumination zones (30a-b); and wherein in particular corrected events are respectively collected only within the same illumination zone (30a-b).

7. A method in accordance with claim 6, wherein the illumination zones (30a-b) differ from one another in at least one of the following illumination properties: reflected light or transmitted light, polarization, intensity, spectrum, angle of incidence, illumination pattern.

8. A method in accordance with any one of the preceding claims, wherein the time interval corresponds to the time duration within which the homogeneous surface (16) is moved through an illumination zone (30a-b) or through the whole field of view (18) of the image sensor (20); and / or wherein the movement taking place during the time interval is specified by a parameterization, measured by means of an additional sensor, or determined from the events.

9. A method in accordance with any one of the preceding claims, wherein the object (14) only moves in one direction (12), in particular uniformly.

10. A method in accordance with any one of the preceding claims, wherein the events are corrected in accordance with the specification Xn = X - v * dT, where X is the position of the pixel element triggering the event, Xn is the new position, v is the speed of the movement, and dT is the time elapsed since the reference point in time.

11. A method in accordance with any one of the preceding claims, wherein the events themselves are evaluated for the correction corresponding to the time that has elapsed since a reference point in time and the movement that has taken place during the elapsed time, without first composing an image therefrom, and the corrected results are only finally collected in an image after the correction, the image then being evaluated with respect to the defects (17).

12. A method in accordance with any one of the preceding claims, wherein the image is evaluated using a process of machine learning, in particular a neural network, as to whether it has defects (17).

13. A method in accordance with any one of the preceding claims, wherein an event has respective coordinate information of the associated pixel element, time information, and / or intensity information.

14. A method in accordance with any one of the preceding claims, wherein the event-based image sensor (20) generates image information having a refresh rate of at least one KHz or even at least ten KHz, and / or wherein a respective pixel element determines when the intensity detected by the pixel element changes and generates an event exactly then, with the event in particular having differential information whether the intensity has decreased or increased.

15. A camera device (10) for recognizing defects (17) on a homogeneous surface (16) of an object (14), in particular a film, during a movement of the object (14) relative to the camera device (10), that has an image sensor (20) for recording an image of the homogeneous surface (16) and has a control and evaluation unit (24) which is configured to evaluate the image as to whether it has defects (17), characterized in that the image sensor (20) is an event-based image sensor having a plurality of pixel elements which detect events of a changing intensity; and in that the control and evaluation unit (24) is configured to record events over a time interval within which the homogeneous surface (16) moves on over at least some pixel elements of the image sensor (20), to correct the events in accordance with a time that has elapsed since a reference point in time and the movement that has taken place during the elapsed time and to collect the corrected events in an image which is then evaluated with respect to the defects (17).