Method for estimating a weight and / or a length of a fish

Abstract

A computer implemented method for estimating a weight of a fish is provided, by using a method comprising: a) capture an image of the fish by, b) transmit the image of the fish (10) to at least one AI model, c) transmit the segmented image to a skeleton algorithm, d) use a selection of the stored pixels and convert them to real coordinates that are positioned along the calculated centre line of the fish, e) remove the edge points, f) sum up all real coordinates to obtain an estimate of a length of the fish.

Description

[0001] METHOD FOR ESTIMATING A WEIGHT AND / OR A LENGTH OF A FISH

[0002] Field for the Invention

[0003] The present invention relates to a computer implemented method for analysing a fish based on a picture captured by a mobile phone or the like so that one can register the species of the fish, weight, length and where it was caught and at what time.

[0004] The invention is primarily meant for outdoor use in nature, in places without mobile coverage and that the picture of the fish can be captured while hanging in the air or lies in a net but not restricted to this alone.

[0005] Background for the Invention

[0006] Today, there are several inventions relating to the measurement of biomass in fish in fish cages, that are not well suited to measuring the biomass of a single fish for which one also has to determine the species. These are usually very expensive and are used in the context of cameras connected to computers and thus unsuited for bringing along on a field trip. An example of such a method and system can be found in the European patent publication EP4008179 A1.

[0007] There are also other known methods that use a combination of manual and digital tools to measure the length of a fish, such as in the US patent US 9,020,416 B2.

[0008] Being able to measure the species, length and weight of a fish by a simple mobile unit with good precision for weight and length while holding the fish by the hand in the air or lying in a net so that the fish is not damaged, have been some of the motivations for this invention.

[0009] EP4008179A1 discloses according to its abstract, a method of determining biomass of at least one aquatic animal, wherein the method comprises detecting a set of key points based on the determined features corresponding to the at least one aquatic animal using a first trained algorithm.

[0010] US2018263231 A1 discloses according to its abstract, a system for fish identification and information that includes: a database accessible through a server via a network; a plurality of fish images stored on the database; a software application on a portable electronic device, where the portable electronic device connects to the database via the network to gain access to the plurality of fish images; and an image capture mechanism integrated into the portable electronic device where the image capturing mechanism is adapted to capture a first image for comparison with images within the plurality of fish images.

[0011] GR1010400B discloses according to a translation of its abstract, a system and method is implemented with the aid of a software which estimates the length and weight of individual fish among a set of fish in real industrial rearing conditions either in tanks or in fish cages.

[0012] Summary of the Invention

[0013] The present invention aims in general to solve at least one but preferably a plurality of the problems that exist in prior art. More specifically, it has been and objective for the invention to develop a more efficient method for analysing fish from an image captured by a mobile computer unit without access to any Internet.

[0014] An objective has been to provide a method for finding the weight of a fish by use of a mobile computer unit by use of an AR camera or similar technology that can convert to 3D real coordinates (z, y, z).

[0015] These objects are achieved by a method as defined in the independent claim 1. Further embodiments of the method are provided by the dependent claims 2 - 10.

[0016] Thus, it is provided a computer implemented method for estimating a length and / or a weight of a fish as well as species of fish, wherein the method comprises the following steps:

[0017] a) capture an image of the fish by use of a camera associated with a mobile computer unit such as a mobile phone or a tablet,

[0018] b) transmit the image of the fish to at least one Al model, wherein the image:

[0019] - is classified, so that it is determined which species is in the image, and - is segmented, so that all pixels of the image that comprise the fish by setting a first value, and remaining pixels are set to a second value, c) transmit the segmented image to a skeleton algorithm to calculate a centre line for the fish, and store all pixels that are located on the calculated centre line, d) use a selection of the stored pixels and convert them to real coordinates that are positioned along the calculated centre line of the fish, e) remove the edge points (end points) of the selected stored pixels and related real coordinates and calculate new edge points with new real coordinates,

[0020] f) sum up (integrate) all real coordinates to obtain an estimate of a length of the fish,

[0021] g) apply empiric length and weight values for the classified species of the fish, as obtained in step b) to estimate the weight for the fish based on the estimated length, as obtained in step f).

[0022] Empiric length and weight values in feature g) can be handled in several ways. In one embodiment, a table can be used, optionally with interpolation. This table can also be adapted for the body thickness of the fish. The table will typically be specific for the species of the fish

[0023] In an example, the estimated weight is given by the following formula:

[0024] W = aLb

[0025] wherein a and b in the formula are set so as to be able determine weight given empirical data, or a power function, and where a and b are variables which depend on species and the individual body thickness of the fish.

[0026] Typically, the values of a and b are stored in tables and are based on data sets of measurements of various species of fish and length and weight relationship.

[0027] The body thickness can be obtained from the segmentation process in step b) and a total area of the fish can also be determined by summing the areas of each segment. From the area one can estimate the body thickness. One can also use a plurality of images of the fish from different angles to better determine a 3D shape and thereby achieve a better estimate of the thickness.

[0028] Further inventive embodiments of the invention are set forth in the independent patent claims.

[0029] Embodiments of the Invention

[0030] The present inventio relates to a computer implemented method for estimating a length and / or a weight of a fish as well as species of fish, wherein the method comprises the following steps: a) capture an image of the fish by use of a camera associated with a mobile computer unit such as a mobile phone or a tablet,

[0031] b) transmit the image of the fish to at least one Al model, wherein the image:

[0032] - is classified, so that it is determined which species is in the image, and - is segmented, so that all pixels of the image that comprise the fish by setting a first value, and remaining pixels are set to a second value, c) transmit the segmented image to a skeleton algorithm to calculate a centre line for the fish, and store all pixels that are located on the calculated centre line,

[0033] d) use a selection of the stored pixels and convert them to real coordinates that are positioned along the calculated centre line of the fish, e) remove the edge points (end points) of the selected stored pixels and related real coordinates and calculate new edge points with new real coordinates,

[0034] f) sum up (integrate) all real coordinates to obtain an estimate of a length of the fish,

[0035] g) apply empiric length and weight values for the classified species of the fish, as obtained in step b) to estimate the weight for the fish based on the estimated length, as obtained in step f).

[0036] An advantage of using real coordinates, or AR, meaning Augmented Reality, is that a layer of digitally added information can be layered on top of an image, that may be directly provided by a camera.

[0037] Real coordinates (the x, y, z coordinates) of a point or a node in relation to a coordinates system overlaid the real world. This is also simply called “3D World Coordinates”.

[0038] An advantage of the method is that it overcomes the problem of points provided as edge values, or end points, of the image of the fish. The edge value problem can arise when AR points (augmented reality point) are to be placed, or real coordinates on the edge points that define the borders of an image or individual such as a fish. The objective of the edge points is that these points should accurately represent the edges or end points of the object., with specific points on important points such as tip and tail end of a fish. To ensure that the points are on the actual edges of the object it is important to obtain precise visual or analytical representations of the object. In the present method step e) the edge points are removed from the stored pixel on the edge points and new edge points are calculated to be certain that the AR points, the real coordinates, do not miss the fish, especially in the depth of the image, when positioned. This means that the fish in the image does not have to rest on a fixed background but that the fish can be depicted while hanging in the air.

[0039] It is also an advantage of the method that it employs empiric length and weight values from a large body of compiled data that provide good estimates for the interrelationship between fish species, length, and weight. In a further embodiment the method comprises the following step e1) after step e) wherein deviating real coordinates, or outliers, such as due to the dorsal fin, are filtered out from the calculated centre line of the fish so that the selected real coordinates are located close to the centre line of the fish. This is an advantage for cases where, for instance, for a picture of a fish where a finger or an entire hand is obstructing the fish on the image when holding the fish. In such cases, the real coordinates could end up on a line on the fish above or below the finger or the hand, and that would result in an incorrect measure of the length of the fish had the real coordinates not been filtered.

[0040] It is also an advantage to employ a skeleton algorithm or topological skeleton, also known as topological reduction, which is a method used in image processing and computer vision to simplify and represent an object into a series of lines or curves. Topological skeletons comprise typically a thin and robust representation of a shape that maintains its geometrical form and structure but eliminates unnecessary details.

[0041] In an embodiment, the method in step g) comprises that the estimated weight is calculated by the following formula:

[0042] W = aLb

[0043] wherein a and b in the formula to determine weight given empirical data, or power function, and where a and b are variables with respect to species and the individual body thickness.

[0044] This has an advantage over known k-factor formulas (Fulton’s condition factor et. 1911) wherein only the k is variable, and the exponent is locked to the third. The empiric formula Wc=kcL3or kc= 100*W / L3is particularly suited for trout and for trout the third power exponent is a very good fit, while the exponent can vary for other species of fish and the thickness of the individual. In an embodiment, the method comprises that the mobile computer unit is not connected to a network. The mobile computer unit can be used independently of the mobile network; everything is performed locally on the mobile unit which is more flexible for the user. It is a known problem with mobile network coverage along the coast and inland, and it would therefore be an advantage that the method does not rely on mobile coverage or network in order to operate.

[0045] In an embodiment, the method comprises an artificial intelligence model (Al-model) for classification trained by pictures of fish, wherein the pictures of fish are cut up and stored in a 4:3 format, and by use of a convolutional neural network (CNN). Small 4:3 pictures and CNN are used to create software that does not take too much memory. There is limited storage capacity on mobile units, and the user friendliness is improved by the use of simple Al models since these are faster. Typical small Al models such as EfficientNet version BO to B3 are suited for use on smaller processors.

[0046] Classification model based on a convolutional neural network (CNN) is a type of deep learning architecture that is particularly suited for identifying patterns in pictures. The model is built by transmitted learning wherein pre trained weights from an established CNN architecture is used as a starting point This provides a faster and more precise model training, since the basic pattern recognition is already built in. The weights are optimised on large and general image data sets and have been finely tuned with specific data that has been collected and that had been manually edited to optimise the learning and the performance of the model. By use of the collected data, that is particularly relevant for fish, the model can be fine-tuned in order to recognize different types of fish in images. Manual editing of the data set will ensure a higher quality and improved precision in classification since the model learns based on clear, relevant, and representative examples.

[0047] An embodiment of the method comprises the Al-model for segmentation trained for segmentation of fish objects by use of a ll-Net and then stores a binary image with “1” pixels that represent the body of the fish, and “0” pixels that represent background. Il-Net is a convolutional neural network (CNN) that is well suited for segmenting images in a simple manner. U-Net++ is a segmentation model that can be used to yield a segmented image of all fishes in the images. U-net++ is an improved version of the well-known ll-Net architecture, that is developed to segment detailed objects from the background of an image. By use of pre-trained weight from a deep learning high performance architecture, the U-Net++ model can achieve a robust ability to learn detailed and precise segmentation patterns with minimal overfit, which is important for images comprising different species of fish and background elements, wherein “1” pixels that represent the body of the fish and “0” pixels that represent the background elements. The Al model is trained by compiled data, a data set, that is edited and optimised for the purpose of segmentation. This data set will provide the model with a good basis for learning to separate the fish from the background and deliver precise segmentations. The segmentation process is therefore adapted to images of fish, and the model delivers images where the fish are clearly marked and separated, which eases further analysis and optionally classification.

[0048] In an embodiment of the method, the Al model is divided into two tasks. These tasks can be performed by one Al model or two Al models, and with current technology for mobile computer units, it will be more efficient to use two models. For a mobile computer unit that is not connected to any net it is an advantage that the Al model is simple and therefore also fast, and there is usually a plurality of processors in a mobile computer unit that can work simultaneously and independent of each other. In an embodiment of the invention, there is a classification mode, Al model 1, that recognizes the image with the fish and classifies into a species of fish, and a segmentation model, Al model 2, that divides the image into meaningful segments to identify fish and that is used further in a skeleton algorithm.

[0049] In an embodiment of the method, the Al model comprises an object and scene recognition model, which is a more advanced Al technology, wherein classification and segmentation are performed using more advanced methods to recognize objects and scenes in images. This makes it possible to search for specific motives or themes such as “sunset” or “fish” and find relevant images in an image base of images captured earlier.

[0050] In an embodiment of the method, the Al model comprises a software that simplifies machine learning for mobile and embedded units, so that they can build intelligent applications that run locally on the mobile unit, without any need for network connectivity or cloud-based processing.

[0051] In an embodiment of the method, the Al model comprises a machine learning framework designed to run machine learning models on mobile and embedded units as well as other resource constrained environments.

[0052] Typically, there will be software libraries that let developers distribute machine learning models on mobile units, including smart phones, tablet and even microcontrollers.

[0053] The advantage of using machine learning models suited for mobile and embedded units, is that these provide high performance and low latency, so that the model can be run efficiently on units with limited processor power and memory. This is achieved by using techniques such as model quantization, which reduces the accuracy of the weights and biases of the model to permit that they can be stored and processed by use of fewer bits. TensorFlow or TensorFlow lite are examples of such machine learning models.

[0054] In an embodiment of the method, the skeleton algorithm is based on a function that is designed to process a binary image of a fish, with “1” pixels that represent the body of the fish, and “0” pixels that represent background, and to identify its centre line from tip to tail, wherein the centre line can serve as a reliable reference for measuring the length of the fish in augmented reality (AR) applications such as real coordinates (x, y, z). It is an advantage that the real coordinates can be placed directly on the image, and directly on the fish, via the camera that has distance measurements or optical remote measurement technique such as LIDAR.

[0055] In an embodiment of the method, the skeleton algorithm in step c) comprises the following steps:

[0056] c1) where the segmented image is transformed to a 2D matrix with x-, y-, and z-coordinates, where the fish is represented by pixels having value “1”, c2) calculate a horizontal centre line by x-iteration of the fish,

[0057] c3) calculate a vertical centre line by y-iteration of the fish,

[0058] c4) determine orientation of the fish based on steps c2) and c3), vertical or horizontal, and calculate a centre line in relation to the orientation of the fish, c5) store the detected centre line points that are on the calculated centre line, wherein step c1 - c5) executes after step b) and before step d).

[0059] The skeleton algorithm is used to measure the length of the fish by selecting a selection of point pixels from the skeleton algorithm and transform these to AR-points or real coordinates. For this reason, the centre line detection of the skeleton algorithm is critical for accurate length measurements.

[0060] The advantage of using AR-points for measurement is that a selection of the centre line points provides an accurate, consistent measurement from the tip to the tail without having to account for body contour or fins. It is also an advantage with flexibility of orientation, and by examine both horizontal and vertical adjustments, it adapts to images of fish in various orientations and ensures robust detection. Thereby a precise orientation independent way to measure the length of the fish in AR application based on the core body centre line. In an embodiment of the method, the new edge points are calculated based on the following formula:

[0061]

[0062] wherein the y is the new edge point for the stored centre line and xi,yi is the closest stored pixel and X2,y2 is the closest stored pixel of the original centre line points. The AR-points, or real coordinates, for the edge points are important for obtaining an accurate measure of the fish. Edge poits are transformed to AR-points, or real coordinates, and if the pixel on the edge point that contains fish causes the AR-point to end up outside the fish, the measurements will be wrong. The advantage of determining the edge points from the closest AR-points is that it enables measurement of the fish when hanging in a room without having the AR edge points ending up far outside the fish, so that the depth measurement of the fish becomes wrong.

[0063] In an embodiment of the method, the mobile computer unit comprises a camera with depth measurements and wherein the mobile computer unit has an AR application installed which makes it possible to insert real coordinates (x, y, z) into the image using a camera with depth measurements.

[0064] In an embodiment of the method, the mobile computer unit comprises a software development kit for small mobile units, such as ARCore or ARkit, to make augmented reality (AAR) more accessible and user friendly for developers and users of mobile units.

[0065] While the embodiments disclosed relate to fish, the invention is not restricted to fish alone. For example, crustaceans such as lobsters can also be processed as disclosed herein. The method can be particularly useful when used to determine features of the tail structure of a lobster, in particular when processing edge points outlined in step e). Thus, the wider field of elongated aquatic animals having a tail or tail fin can be handled by the method disclosed herein.

[0066] Preferred embodiments of the invention will in the following be disclosed in more details with references to the following figures, wherein:

[0067] - Fig. 1 shows a flowchart of a method for estimating the weight of a fish. - Fig. 2 shows a flowchart of the sequences obtaining the length of a fish. - Fig. 3 shows an image of a fish, a segmentation image (binary) and three different skeleton algorithm images. - Fig. 4 shows a flowchart to calculate the weight of a fish by use of a mobile telephone.

[0068] - Fig. 5 shows an overview of a method for calculating new edge values for a topological skeleton of a fish.

[0069] Fig. 1 shows a flowchart of an embodiment of the invention wherein an AR image has been captured of a fish from the side 10, and wherein the AR fish image 10 is transmitted to an artificial intelligence (Al) model, which consists of two parts. A first part of the Al model classifies the fish image to a fish species 30, and a second part segments the image down to M number of pixels, typically 400 x 40020. Different resolutions can be employed. Low resolutions provide for faster calculations and better performance on low powered devices, while high resolutions require more computing power.

[0070] Next the segmented image 20 is transmitted to a skeleton algorithm 40, which determines the orientation of the fish and forms a topological skeleton of the fish from tip to tail in a binary 2D image. In the next step, N number of AR points are laid out, for instance seven, along the points from the skeleton algorithm from tail to tip of the fish 50. In the next step, the edge points are removed, and new edge points are calculated 60. Next, these AR points / real coordinates) with the next AR edge points are used to calculate the length of the fish 70. Then the calculated length and fish species, as determined in the classification, are used to estimate the weight of the fish 80. The empirical formula is given as

[0071] W = aLb,

[0072] wherein W = the weight, and L = the calculated length, and wherein a and b vary in relationship to the species of the fish and the thickness of the individual, as thick, medium, or thin.

[0073] Fig. 2 shows a flowchart from an image of a fish 10 to AR points for calculating the length of the fish 70. The image of the fish is of good resolution, many pixels, which in next step is segmented down to a binary 2D image 20 of the fish by for instance 400 x 400 pixels. In the binary 2D image 20 of the fish, everything that is fish is set to one value, and everything that is not fish is set to another value from the segmentation process performed by an Al model. The binary image of the fish 20 is used in the next step to form a topological skeleton of the fish 40 by use of a skeleton algorithm. The topological skeleton from the skeleton algorithm is used to place N number of AR points (real coordinates), here shown as seven points. It is shown on the image that one end point, AR edge point, hits the tip of the tail, which will yield the wrong measurement of the length of the fish 50. This error is corrected in the next step, where the AR edge points are removed and new AR edge points are overlaid, thus the last end point is now on between the two tips of the tail which yields a more exact length of the fish 60. In the last step, AR points 70 are used to calculate the length of the fish on the image of the fish 10.

[0074] Fig. 3 shows fish images 10 in first column and in second column a segmented fish image 20 (binary 2D image) of the fish images 10 in the first column used in a skeleton algorithm to provide a topological skeleton. In the next three columns, three different skeleton algorithms are shown:

[0075] In the third column, Zhang’s method (a fast parallel algorithm for thinning digital patterns, T. Y. Zhang and C. Y. Suen, Communications of the ACM, March 1984, Volume 27, Number 3.)

[0076] In the fourth column, Lee’s method (T.-C. Lee, R.L. Kashyap and C -N. Chu, Building skeleton models via 3-D medial surface / axis thinning algorithms. Computer Vision, Graphics, and Image Processing, 56(6):462-478, 1994.).

[0077] In the fifth column, an in-house developed (Fiskher) algorithm that is tailor made for fish and that also takes orientation of the fish into account. It is much more precise for use in providing a topological skeleton of the fish and it does not take fins and other deviations into account as the two preceding algorithms do.

[0078] A topological skeleton can be calculated by algorithms that iteratively thin out an image or a form by removing pixels or voxels that are not part of the skeleton. This results in a representation of the object as a series of points that represent the centres of the original forms or lines.

[0079] Topological skeletons can be used in many applications within image processing and computer vision, such as analysing biological tissue, recognizing objects and patterns, and segmenting images.

[0080] It can also be useful in reducing the amount of data required to represent an object, which can be beneficial in applications with limited storage capacity or processing power.

[0081] A part of the invention has been to create a topological skeleton based on a binary image provided by the segmentation model.

[0082] The skeleton algorithm, the topological skeleton function, is designed to process a binary image of a fish, with “1” pixels that represent the body of the fish, and “0” pixels as background, and to identify its centre line from tip to tail. This centre line can then serve as a reliable reference for measurements of the length of the fish in augmented reality (AR) applications, that is in real coordinates and then results in a unit of length such as centimetre or inch.

[0083] Fig. 4 shows a flowchart to determine the weight of a fish by use of a mobile phone (smart phone) using an application (app) that is installed on the mobile phone. Start by opening the application on the mobile phone, then in step a) capture an image of the fish with a camera with AR-function and obtain directly the species as well as length and weight of the fish.

[0084] In the next step b) the image is sent to two Al models; the left branch executes steps b1) which comprises segmenting the image, step c) which comprises determining the centre line, step d) which comprises placing multiple, for instance 7, AR points along the centre line, step e) comprises removing edge points, and step f) comprises summing up (integrating) all real coordinates to obtain an estimate of a length of the fish; while the right branch executes steps b2) of classifying the fish in the image. Combining the outputs in b1) and f), step g) comprises using a table with known values for length and weight to estimate the weight of the fish; and thereby provide data including one, some, or all of the species, length, and weight of the fish. Fig. 5 shows how the two new AR edge points NINY and NXNY can be calculated from the two closest AR points, here given by the coordinates (xi,yi) and (X2y2) as a general example of how it can be performed.

[0085] The invention is now disclosed with several non-limiting embodiments. A person skilled in the art will understand that a system according to the invention can comprise several combinations and variations based on the provided examples.

[0086]

[0087]

Claims

CLAIMS1. A computer implemented method for estimating a weight and / or a length of a fish, wherein the method comprises the following steps:a) capture an image of the fish by use of a camera associated with a mobile computer unit such as a mobile phone or a tablet,b) transmit the image of the fish (10) to at least one Al model, wherein the image (10):- is classified (20), so that it is determined which species is in the image, and- is segmented(30), so that all pixels of the image that comprise the fish by setting a first value, and remaining pixels are set to a second value, c) transmit the segmented image to a skeleton algorithm to calculate a centre line (40) for the fish, and store all pixels that are located on the calculated centre line (40),d) use a selection of the stored pixels (NI,2,3,...X) and convert them to real coordinates (Xi,2,3...x, YI,2,3...X ,Zi,2,3...x) that are positioned along the calculated centre line of the fish (40),e) remove the edge points (end points) of the selected stored pixels (Ni og Nx) and related real coordinates (Xi,Yi„Zi og Xx, YX,ZX) and calculate new edge points (N-iNYog NXNY) with new real coordinates (XINY.YINY, ZINY and XXNY, YXNY, ZXNY),f) sum up (integrate) all real coordinates (XINY,2,3...XNY, YINY,2,3...XNY ,ZINY,2,3...XNY) to obtain an estimate of a length of the fish,g) apply empiric length and weight values for the classified species of the fish, as obtained in step b) to estimate the weight for the fish based on the estimated length, as obtained in step f).

2. The computer implemented method according to claim 1, characterised in that the method further comprises the following step: e1) filtering out any real coordinate outliers from the calculated centre line of the fish (40) so that the selected real coordinates (x, y, z) are located close to the centre line of the fish, wherein the step e1) is performed after step e) and before step f).

3. The computer implemented method according to one of the above claims, characterised in that the estimated weight in step g) is calculated by the following formula:W = aLbwherein a and b in the formula to determine weight given empirical data, or power function, and where a and b are variables with respect to species and the individual body thickness (thick, medium, and thin).

4. The computer implemented method according to one of the above claims, characterised in that the mobile computer unit is not connected to a network.

5. The computer implemented method according to one of the above claims, characterised in that the Al-model in step b) for classification is trained by images of fish that are cut and stored in a 4:3 format, by use of a convolutional neural network.

6. The computer implemented method according to one of the above claims, characterised in that the Al-model in step b) for segmentation is trained for segmenting fish objects by use of a ll-Net and forms a binary image with “1” pixels representing the body of the fish and “0” pixels representing background.

7. The computer implemented method according to one of the above claims, characterised in that the skeleton algorithm in step c) is based on a function designed to process a binary image of a fish with “1” pixels representing the body of the fish and “0” pixels representing background and to identify its centre line from tip to tail, wherein the centre line can serve as a reliable reference to measure the length of the fish in augmented reality (AR) applications such as real coordinates (x, y, z).

8. The computer implemented method according to one of the above claims, characterised in that the skeleton algorithm in step c) further comprises the steps:c1) wherein the segmented image is transformed to a 2D matrix with x- and y-coordinates, wherein the shape of the fish is represented by pixels having value “1”,c2) calculate a horizontal centre line by x-iterations of the fish,c3) calculate a vertical centre line by y-iterations,c4) determine an orientation of the fish based on steps c2) and c3), vertical or horizontal, and calculate a centre line in relation to the orientation of the fish,c5) store the detected centre line points that are on the calculated centre line, wherein steps c1) - c5) are performed after step b) and before step d).

9. The computer implemented method according to one of the above claims, characterised in that the new edge points in step e) are calculated by the following formula:wherein the y is the new edge point on the stored centre line, and wherein xi.yi is the closest stored pixel and X2,y2 is the closest stored pixel of the original centre line edge points.

10. The computer implemented method according to one of the above claims, characterised in that the mobile computer unit has an augmented reality (AR) application installed, which makes it possible to insert real coordinates (x, y, z) into the image using a camera having depthmeasurements.

Images