Ai-based analysis of a video stream or of photographs for calculating time-variable coordinates and movement angles of the joint points of four-legged animals

Abstract

The invention relates to a computer-assisted method for identifying joint positions and / or angular movements of joints, in particular for analyzing the gait of living beings, in particular four-legged animals, in particular horses, ponies, camels and / or dogs, comprising providing at least one image, image stream and / or video stream showing a living being, in particular a four-legged animal, in particular from the side, in particular while moving and from the side in relation to a movement direction of the living being; and determining at least one joint point and / or anchor point on the living being in the image, image stream and / or video stream by means of a machine-learning model, in particular on the basis of a movement-anatomy model of the living being, in particular a species-specific and / or genus-specific movement-anatomy model.

Description

[0001] AI-based analysis of a video stream or photos to calculate time-varying coordinates and movement angles of the joint points of quadrupeds

[0002] - Patent application -

[0003] Field of the invention

[0004] The present invention relates to the field of analyzing anatomical, particularly anatomical-dynamic, properties of living beings. The invention is particularly suitable for use in horses (e.g., racehorses, dressage horses), ponies, camels, and dogs, but also donkeys, cats, monkeys, and tapirs. A quantitatively highly accurate and, based on this, qualitatively accessible and understandable analysis of the gait mechanics of the examined living beings is provided. Possible commercial applications include (preventive and curative) veterinary medicine, also for improving performance, e.g., through the treatment and elimination of previously undetected anatomical and / or dynamic asymmetries in the movement of individuals, as well as, for example, an improved market value determination of the animals through more precise knowledge of dynamic performance characteristics.

[0005] 1

[0006] REPLACEMENT SHEET (RULE 26) Technical Background

[0007] Animals play a wide variety of roles in modern society. They bring joy to people. Both animal husbandry and animal use are widespread. In many cases, animals are important commercial goods and / or competitive athletes. Riding horses are an example. Another example is horse betting or the sale of a dog, donkey, or camel.

[0008] Often, the animals' anatomical characteristics are thoroughly examined. This is usually done by visual inspection, particularly visual examination and / or examination by an expert in the respective species and / or genus. This could be a horse enthusiast with extensive animal experience, an experienced camel dealer, or even a veterinarian.

[0009] In such an examination, dynamic characteristics (such as severe behavioral abnormalities during movement) can also be taken into account, within certain limits. However, such analyses are often highly imprecise. Even though experts often like to consider themselves to be almost entirely competent, i.e., animal-related, they are based on "rough guesswork." In particular, minor movement abnormalities can very easily go undetected. In the worst case, these abnormalities are self-reinforcing, so that what initially appears to be only a minor problem in an animal develops into a major problem later on. Due to the lack of knowledge, the potential for compensatory and / or veterinary measures, including those that increase the animal's performance, remains untapped. The state of the art methods do not allow the animals' potential to be maximized.The situation is therefore highly unsatisfactory.

[0010] The disadvantages of the prior art are overcome by the computer-implemented method according to claim 1 for detecting joint positions and / or angular movements of joints. Description of the invention

[0011] The present invention provides a method having the features of claim 1. Further advantageous embodiments are specified in the subclaims.

[0012] The problem is solved by a computer-implemented method for recording joint positions and / or angular movements of joints in living beings. The method can be used in particular for analyzing gait mechanics. The method can be used particularly in four-legged animals, for example, horses, ponies, camels, and / or dogs.

[0013] The method can comprise providing at least one image, image stream, and / or video stream showing a living being, in particular a quadruped. This can be done, in particular, from a lateral perspective. For example, corresponding image and / or video material is recorded using a corresponding camera positioned laterally, or multiple cameras. In particular, the running living being can be recorded, in particular from a perspective lateral to its running direction. Various further developments, in particular also concerning the beneficial use of multiple cameras, will be discussed later in the context of further developments and / or embodiments of the invention.

[0014] The method may further comprise determining at least one joint point and / or anchor point on the living being in the image, image stream, and / or video stream using a machine learning model. In particular, a movement anatomy model of the living being may be used. In particular, this may be a species- and / or genus-specific movement anatomy model, for example, for a horse (or for a specific type of horse, or for a higher-level classification such as perissodactyla). Such a model may, for example, include specific joint and / or anchor points and / or angles at the joints. Certain constraints may be included in some models.

[0015] The invention enables extremely precise quantitative determination of joint positions, joint angles, ranges of motion, and amplitudes. Gait analysis is thus possible with previously unknown precision. Effects such as asymmetries become visible, which even the most experienced experts cannot diagnose with the naked eye. Additional anatomical data, distances traveled, speeds, and accelerations can be precisely determined and taken into account. Precision is improved by using a movement anatomy model, particularly one that is tailored to the respective living being (e.g., genus / species / breed). Such determination can, for example, also be performed by Kl in a prior step, so that a particularly suitable anatomy model can then be selected for determining joint positions.The invention allows important conclusions to be drawn about the anatomy and dynamics of a living being from data that would otherwise be inaccessible.

[0016] Recording two image streams allows for observation and analysis of both sides of the animal. Asymmetries can also be analyzed and more precisely determined using artificial intelligence (AI). An AI model can, for example, include an artificial neural network (ANN or ANN). Different and / or separate AI models can be used for the sides of the animal. However, it is particularly advantageous to use a common AI model for both sides of the animal, with one of the two image streams mirrored vertically. Figuratively speaking, "both animals are then walking in the same direction."

[0017] This vertical mirroring can be performed explicitly on the data stream or implicitly. Explicitly, a data area is actually created with a mirrored video stream. Implicit methods utilize, for example, a specific reading direction of a raw video data stream, or a special intermediate interface, or the like. Those skilled in the art are familiar with numerous ways to technically implement this effective mirroring for further processing of the data with the AI. The invention is by no means limited in this respect.

[0018] A side view is any view of an animal where its side is visible. The animal doesn't necessarily have to be oriented perpendicular to the line of sight / camera direction and / or walking. Any perspective that doesn't show the animal exactly from the front or back can be considered a side view.

[0019] Detected joint positions can be graphically displayed, for example, in an output video stream or a corresponding output image. The same applies to angles.

[0020] Numerous plots can be created and made available to the user for detailed analysis of the animal. Pie charts are particularly clear, providing angle ranges (minimum angle, maximum angle) of the results generated by the KL (for both sides, left and right). This allows asymmetries to be quickly and clearly identified and quantified.

[0021] For a better understanding of the embodiments, reference is made to the figures and the associated figure descriptions. The use of time stamps, in particular universal Unix time stamps, on the data streams has excellent properties for later synchronization. This enables the desired analyses and insights to be obtained, particularly in the case of incomplete video material (deletions, technical failures, etc.), which would be difficult to re-obtain. With multiple cameras, 3D analyses (e.g., several / numerous angle-shifted cameras) are also possible with high precision, made possible, among other things, by the ability to rely on precise universal time synchronization (from the source, i.e., not subsequently introduced by video comparison—such time stamp sources often prove to be too imprecise).

[0022] 3D analyses, which are more complex, are particularly popular in competitive equestrian sports. Angle analyses can also be generalized to three dimensions.

[0023] Accelerations can be included in the analysis (2D and / or 3D), including angular accelerations. These are important performance variables in sporting animals and are almost impossible to detect with the naked eye and / or are subject to highly subjective perceptions / observer cognitive biases.

[0024] It may be advisable to perform the analysis twice: first, a rough preliminary analysis (e.g., using image / video material compliant with the High Definition HD video standard), followed by a more detailed analysis (e.g., based on high-resolution 4K video material). With this multi-pass strategy, the results of the first pass are available quickly, eliminating the need for the user to wait. The subsequent high-quality pass (e.g., 4K) then delivers more precise results. Differences between the individual passes can also be used a) to estimate errors and b) to detect the occurrence of analysis errors.

[0025] In a further development of the invention, acoustic signals of the animal (movement noises) are recorded and evaluated.

[0026] Preferably, a reference object is placed in the image, which the AI ​​recognizes as a reference object and uses to recognize dimensions and / or improve the conversion accuracy of scales. This enables automatic geometric compensation. It also enables dynamic consideration of a camera distance. This is preferably a reference cuboid or a reference rectangle (a reference square is also possible). Experiments have shown that the algorithm becomes particularly stable when additional points with a predefined diameter and / or predefined distances from one another are placed on the reference object (and taken into account). Training of the analysis AI is preferably supplemented, as experiments have shown, by AI techniques of negative learning.In this type of "unlearning," negative examples, especially those that occur during regular operation but are identified as faulty, are used as "negative training data." The AI ​​is thus trained to avoid this behavior in the future. This technique produces a particularly high level of convergence in the application and thus effectively eliminates errors, which could not be achieved with pure "positive learning."

[0027] The invention can be commercially exploited in many ways. Examples include (but are not limited to): new insights in the animal trade and / or performance sports, sports betting, improving performance and health, and exploiting the potential of animals, e.g., through compensatory measures and / or veterinary treatment. The invention provides precise insights that would otherwise remain hidden. The invention does not intend to limit itself to the numerous successive exploitation possibilities of these insights.

[0028] All features disclosed in connection with corresponding methods can be used in connection with the devices, and vice versa. The various features disclosed herein, including those of various embodiments, can be combined with one another.

[0029] The embodiments described schematically here can be further developed by numerous details of the invention, in particular by those already described above.

[0030] Although some aspects have been described in the context of a device, it is clear that these aspects also represent a description of the corresponding method, with a block or device corresponding to a method step or a function of a method step. Analogously, aspects described in the context of a method step also represent a description of a corresponding block, element, or property of a corresponding device.

[0031] Embodiments of the invention may be implemented in a computer system. The computer system may be a local computing device (e.g., personal computer, laptop, tablet computer, or mobile phone) having one or more processors and one or more memory devices, or may be a distributed computing system (e.g., a cloud computing system having one or more processors or one or more memory devices distributed at different locations, for example, at a local client and / or one or more remote server farms and / or data centers). The computer system may comprise any circuitry or combination of circuits. In one embodiment, the computer system may comprise one or more processors, which may be of any type.As used herein, processor may mean any type of computing circuit, such as, but not limited to, a microprocessor, a microcontroller, a complex instruction set microprocessor (CISC), a reduced instruction set microprocessor (RISC), a very long instruction word (VLIW) microprocessor, a graphics processor, a digital signal processor (DSP), a multi-core processor, a field-programmable gate array (FPGA), or any other type of processor or processing circuit. Other types of circuitry that may be included in the computer system may be a purpose-built circuit, an application-specific integrated circuit (ASIC), or the like, such as one or more circuits (e.g., a communications circuit) for use with wireless devices such asMobile phones, tablet computers, laptop computers, two-way radios, and similar electronic systems. The computer system may include one or more storage devices, which may include one or more storage elements suitable for the particular application, such as main memory in the form of random access memory (RAM), one or more hard disks, and / or one or more drives handling removable media such as CDs, flash memory cards, DVDs, and the like. The computer system may also include a display device, one or more speakers, and a keyboard and / or controller, which may include a mouse, trackball, touchscreen, voice recognition device, or any other device that allows a system user to input information to and receive information from the computer system.

[0032] Some or all of the method steps may be performed by (or using) a hardware device, such as a processor, a microprocessor, a programmable computer, or an electronic circuit. In some embodiments, one or more of the key method steps may be performed by such a device.

[0033] Depending on specific implementation requirements, embodiments of the invention may be implemented in hardware or software. The implementation may be performed using a non-volatile storage medium, such as a digital storage medium, such as a floppy disk, DVD, Blu-ray, CD, ROM, PROM and EPROM, EEPROM, or FLASH memory, on which electronically readable control signals are stored that interact (or can interact) with a programmable computer system to perform the respective method. Therefore, the digital storage medium may be computer-readable.

[0034] Some embodiments according to the invention comprise a data carrier with electronically readable control signals that can interact with a programmable computer system so that one of the methods described herein is carried out.

[0035] In general, embodiments of the present invention can be implemented as a computer program product with program code, wherein the program code is effective for executing one of the methods when the computer program product is running on a computer. The program code can, for example, be stored on a machine-readable medium.

[0036] Further embodiments include the computer program for carrying out one of the methods described herein, which is stored on a machine-readable carrier.

[0037] In other words, one embodiment of the present invention is therefore a computer program having a program code for carrying out one of the methods described herein when the computer program runs on a computer.

[0038] A further embodiment of the present invention is therefore a storage medium (or a data carrier or a computer-readable medium) comprising a computer program stored thereon for performing one of the methods described herein when executed by a processor. The data carrier, the digital storage medium, or the recorded medium is typically tangible and / or non-seamless. A further embodiment of the present invention is an apparatus as described herein, comprising a processor and the storage medium.

[0039] A further embodiment of the invention is therefore a data stream or signal sequence representing the computer program for carrying out one of the methods described herein. The data stream or signal sequence can, for example, be configured to be transmitted via a data communication connection, for example, via the Internet.

[0040] A further embodiment comprises a processing means, for example a computer or a programmable logic device, configured or adapted to carry out any of the methods described herein.

[0041] A further embodiment comprises a computer on which the computer program for carrying out one of the methods described herein is installed. A further embodiment according to the invention comprises a device or system configured to transmit (e.g., electronically or optically) a computer program for carrying out one of the methods described herein to a recipient. The recipient may, for example, be a computer, a mobile device, a storage device, or the like. The device or system may, for example, comprise a file server for transmitting the computer program to the recipient.

[0042] In some embodiments, a programmable logic device (e.g., a field-programmable gate array, FPGA) may be used to perform some or all of the functionality of the methods described herein. In some embodiments, a field-programmable gate array may cooperate with a microprocessor to perform any of the methods described herein. In general, the methods are preferably performed by any hardware device.

[0043] Embodiments may be based on using a machine learning model (or multiple machine learning models). Machine learning may refer to algorithms and statistical models that computer systems can use to perform a specific task without using explicit instructions, rather than relying on models and inference. For example, instead of a rule-based transformation of data, machine learning may use a transformation of data that can be derived from an analysis of historical and / or training data. For example, the content of images may be analyzed using a machine learning model or using a machine learning algorithm. In order for the machine learning model to analyze the content of an image, the machine learning model may be trained using training images as input and training content information as output.By training the machine learning model with a large number of training images and / or training sequences (e.g., words or sentences) and associated training content information (e.g., labels or annotations), the machine learning model “learns” to recognize the content of the images, so that the content of images not included in the training data can be recognized using the machine learning model. The same principle can also be used for other types of sensor data: By training a machine learning model using training sensor data and a desired output, the machine learning model “learns” a conversion between the sensor data and the output, which can be used to provide an output based on non-training sensor data provided to the machine learning model. The provided data (e.g.,Sensor data, metadata and / or image data) can be preprocessed to obtain a feature vector, which is used as input for the machine learning model.

[0044] Machine learning models can be trained using training input data. The examples above use a training method called "supervised learning." In supervised learning, the machine learning model is trained using a plurality of training samples, where each sample may include a plurality of input data values ​​and a plurality of desired output values, i.e., each training sample is associated with a desired output value. By specifying both training samples and desired output values, the machine learning model "learns" which output value to provide based on an input sample that is similar to the samples provided during training. In addition to supervised learning, semi-supervised learning can also be used. In semi-supervised learning, some of the training samples lack a desired output value.Supervised learning can be based on a supervised learning algorithm (e.g., a classification algorithm, a regression algorithm, or a similarity learning algorithm).

[0045] Classification algorithms can be used when the outputs are restricted to a limited set of values ​​(categorical variables), i.e., the input is classified as one of the limited set of values. Regression algorithms can be used when the outputs indicate any numerical value (within a range). Similarity learning algorithms can be similar to both classification and regression algorithms, but are based on learning from examples using a similarity function that measures how similar or related two objects are. In addition to supervised learning or semi-supervised learning, unsupervised learning can be used to train the machine learning model. In unsupervised learning, (only) input data may be provided, and an unsupervised learning algorithm can be used to find a structure in the input data (e.g.,Clustering (by grouping or clustering the input data, finding commonalities in the data). Clustering is the assignment of input data comprising a plurality of input values ​​into subsets (clusters) such that input values ​​within the same cluster are similar according to one or more (predefined) similarity criteria, while they are dissimilar to input values ​​contained in other clusters.

[0046] Reinforcement learning is a third group of machine learning algorithms. In other words, reinforcement learning can be used to train the machine learning model. In reinforcement learning, one or more software agents are trained to perform actions in an environment. A reward is calculated based on the actions performed. Reinforcement learning is based on training the one or more software agents to select actions in such a way that the cumulative reward is increased, resulting in software agents that become better at the task given to them (as evidenced by increasing rewards).

[0047] Furthermore, some techniques can be applied to some of the machine learning algorithms. For example, feature learning can be used. In other words, the machine learning model can be trained at least partially using feature learning, and / or the machine learning algorithm can include a feature learning component. Feature learning algorithms, called representation learning algorithms, can preserve the information in their input but transform it in such a way that it becomes useful, often as a preprocessing stage before performing classification or prediction. Feature learning can be based on, for example, principal component analysis or cluster analysis.

[0048] In some examples, anomaly detection (i.e., outlier detection) may be used, which aims to provide identification of input values ​​that are suspicious because they differ significantly from the majority of input and training data. In other words, the machine learning model may be trained at least partially using anomaly detection, and / or the machine learning algorithm may include an anomaly detection component.

[0049] In some examples, the machine learning algorithm may use a decision tree as a prediction model. In other words, the machine learning model may be based on a decision tree. In a decision tree, the observations about an object (e.g., a set of input values) may be represented by the branches of the decision tree, and an output value corresponding to the object may be represented by the leaves of the decision tree. Decision trees can support both discrete values ​​and continuous values ​​as output values. When discrete values ​​are used, the decision tree may be called a classification tree; when continuous values ​​are used, the decision tree may be called a regression tree.

[0050] Association rules are another technique that can be used in machine learning algorithms. In other words, the machine learning model can be based on one or more association rules. Association rules are created by identifying relationships between variables in large data sets. The machine learning algorithm can identify and / or use one or more relationship rules that represent the knowledge derived from the data. The rules can be used, for example, to store, manipulate, or apply the knowledge.

[0051] Machine learning algorithms are typically based on a machine learning model. In other words, the term "machine learning algorithm" may refer to a set of instructions that can be used to create, train, or use a machine learning model. The term "machine learning model" may refer to a data structure and / or a set of rules that represents the learned knowledge (e.g., based on the training performed by the machine learning algorithm). In embodiments, the use of a machine learning algorithm may imply the use of an underlying machine learning model (or a plurality of underlying machine learning models). The use of a machine learning model may imply that the machine learning model and / or the data structure / set of rules that is / are the machine learning model is trained by a machine learning algorithm.

[0052] For example, the machine learning model can be an artificial neural network (ANN). ANNs are systems inspired by biological neural networks, such as those found in a retina or brain. ANNs comprise a plurality of interconnected nodes and a plurality of connections, called edges, between the nodes. There are typically three types of nodes: input nodes that receive input values, hidden nodes that are connected (only) to other nodes, and output nodes that provide output values. Each node can represent an artificial neuron. Each edge can send information from one node to another. The output of a node can be defined as a (nonlinear) function of the inputs (e.g., the sum of its inputs). The inputs of a node can be used in the function based on a "weight" of the edge or the node providing the input.The weight of nodes and / or edges can be adjusted during the learning process. In other words, training an artificial neural network may involve adjusting the weights of the nodes and / or edges of the artificial neural network, i.e., to achieve a desired output for a given input.

[0053] Alternatively, the machine learning model can be a support vector machine, a random forest model, or a gradient boosting model. Support vector machines (i.e., support vector networks) are supervised learning models with associated learning algorithms that can be used to analyze data (e.g., in a classification or regression analysis). Support vector machines can be trained by providing an input with a plurality of training input values ​​belonging to one of two categories. The support vector machine can be trained to assign a new input value to one of the two categories. Alternatively, the machine learning model can be a Bayesian network, which is a probabilistic directed acyclic graphical model. A Bayesian network can represent a set of random variables and their conditional dependencies using a directed acyclic graph.Alternatively, the machine learning model can be based on a genetic algorithm, which is a search algorithm and heuristic technique that mimics the process of natural selection.

[0054] Character list

[0055] The present invention will be explained in more detail below with reference to the exemplary embodiments shown in the schematic figures of the drawings. In the drawings:

[0056] Fig. 1 a horse with a graphic representation of a movement anatomy model,

[0057] Fig. 2 a horse with joint / anchor points P1 - P15 and a graphic representation of a movement anatomy model,

[0058] Fig. 3 a horse with 12 joint / anchor points including definitions,

[0059] Fig. 4 a horse with 8 exemplary joint angles including definitions,

[0060] Fig. 5 is an overview and / or flow chart of an embodiment of the present invention,

[0061] Fig. 6 a sketch to illustrate the two-sided recording of an animal, including mirroring of a video stream,

[0062] Fig. 7 exemplary results relating to a horse’s fetlock joint (example, extract),

[0063] Fig. 8 exemplary angles of the hind legs of a horse in walk,

[0064] Fig. 9 and 10 exemplary pie charts for graphic angle analysis, which quickly provide the observer with the relevant insights, especially in the qualitative and quantitative analysis of asymmetries, and thus enable a more detailed study of the particularly relevant joints and movement patterns,

[0065] Fig. 11 an exemplary movement (vertical movement of the right hip) in detail, individual runs and statistically averaged, from the “right hand” perspective.

[0066] In all figures, identical or functionally identical elements and devices have been provided with the same reference numerals, unless otherwise stated.

[0067] Description of the Drawings Figure 1 shows a schematic representation of a horse R10. It also shows a possible embodiment of a movement anatomy model. Black dots indicate joint points and / or other anchor points. For corresponding definitions of the points shown, please refer to Figure 3. Some example angles are also shown.

[0068] For the horse in Figure 2, a model with 14 joint and / or anchor points is used. Using one embodiment of the present invention, the points are recognized, for example, from a video stream in which the horse in Figure 2 is running using a machine learning method / model / Kl. Angle determination can also be performed using the Kl (increased precision); alternatively, it can be performed using conventional mathematical methods from the determined points.

[0069] In Figure 3, some of the points are named and defined. ("P" stands for point. P3 corresponds to point 3.) Similarly, some corresponding angles are named and defined in Figure 4.

[0070] Figure 5 shows a schematic flow diagram for a preferred embodiment of the present invention. A horse is filmed, for example, when it runs from left to right; or additionally from right to left afterward; or two such video streams are recorded by cameras positioned on both sides (e.g., simultaneously). There can also be more than two cameras and they can be positioned at different angles. (This is particularly useful for an alternative embodiment using 3D anatomical models.)

[0071] The two video streams (or alternatively image streams) from both sides of the horse can, for example, be used as input to a machine learning method / model, e.g., an artificial neural network. Two methods / models, e.g., one for each direction, can also be used (higher precision, provided the networks are sufficiently trained). Preferably, however, a model is used, with one of the two image streams being mirrored (directly or indirectly) (along a vertical axis). This way, the horse "runs" in the same direction in the video streams (although these video recordings correspond to the two different sides of the horse). The precision is better than in the other cases mentioned, and "outliers" (joint points that suddenly and unexpectedly "jump," i.e., discontinuously change their location) are successfully prevented.Such a mirroring is illustrated by way of example in Figure 6: A horse runs from right to left (top) and is filmed. It then runs (afterward / simultaneously) from left to right: It either runs in the opposite direction or is (preferably) filmed simultaneously from the opposite direction. One of the two video / image data streams is then mirrored vertically (technically directly or indirectly), shown here in Figure 6 is the stream of the lower horse (also possible: reversed). The two data streams, where the animals now appear to be running in the "same direction," are preferably passed to a common / unified ML model (for the purpose of analyzing the video data streams; the same applies to training data for effective training of the AI).

[0072] The videos can be time-stamped in the recording tool (upper part of Figure 5), in particular using universal time-stamps, e.g., Unix time-stamps. This allows the material (even incomplete video material) to be temporally correlated later with the highest precision and thus evaluated using the method of the invention. Synchronization can occur, for example, in a synchronization module R30. A storage module / storage medium R40 can store and manage the video streams including time codes. A (Kl-supported) automatic classification of the living being / animal can be provided. For example, a horse or a camel is recognized in this way. Optionally, it can also be recognized whether the animal is mounted or not. Based on the findings, a suitable movement anatomy model can be selected. For example, if a rider is recognized, a "model with rider" is selected.Or a rider model suitable for the rider is selected and combined with a specially designed movement anatomy model of the animal that is ridden by the rider.

[0073] The analysis module can be considered the "heart" of the present embodiment of the invention, without being limiting the invention. Here, the points and angles are determined, possibly even in 3D, based on the selected / determined movement anatomy model and the video streams (possibly including the direction of travel / vertical reflection of a stream). This enables the comprehensive evaluations provided by the output module A90.

[0074] Prior to this, a correlation plausibility module A20 can check the analysis for plausibility. If the plausibility is negative, the material can be discarded and the analysis repeated. In particular, if the plausibility is negative, the material can be forwarded to the AI-based training module A40 for "unlearning."

[0075] It has been shown that such "unlearning" of undesired results is particularly suitable and (quickly) achieves better results and training of the AI ​​network than positive reinforcement alone. Experts are familiar with suitable corresponding techniques in the fields of machine learning and artificial intelligence.

[0076] With geometric correction A50, the points, for example, P1 to P14, are geometrically aligned, and the movement data is modified accordingly. In particular, if an animal runs from left to right once and right to left once, it is not automatically guaranteed that the same camera distance is maintained in both runs, i.e., that the distance between the camera and the object is the same. A simple variant of geometric correction can include methods based on the ray theorem to correct for this difference.

[0077] In the illustrated embodiment, the compared data is transferred to a statistical module A70 and the task module A90. The statistical module compares, in particular, with other living beings / animals / objects of the same class (e.g., camel or tapir) and checks for anomalies. Here, too, the performance of the analysis tool A10 can be further improved by reporting anomalies from the statistical module A70 to the training module A40 (which can be used in a gradual "unlearning" of the analysis class).

[0078] The output module can, for example, provide reports, videos, images, plots, diagrams, etc. Videos and images can include those from the input video stream, in which the fitted movement anatomy model is also clearly displayed. Numerous plots are possible, which depict the temporal progression of certain quantities. Asymmetry ranking is also possible. Here, behaviors that are conspicuous due to asymmetry are clearly presented to the user, hierarchically ordered. Thus, the system starts with the "largest" or (relatively) "most conspicuous" behaviors, without the user having to first view or carefully search through vast amounts of material.

[0079] By calculating average values ​​over several cycles (walking / running cycles of the animal) and corresponding error analysis, deviations (even the smallest) can be determined to be statistically significant.

[0080] A particularly clear and user-friendly presentation of the information can be achieved in the form of pie charts for angle ranges. The angle ranges covered can be highlighted in color. The left and right halves display the angle range for the respective side (left side, right side) of the animal. This allows asymmetries to be identified particularly quickly, possibly already tested for statistical significance by averaging. Additional graphical indicators for asymmetries can be provided for the pie charts, for example, color-coded representations for asymmetries that exceed a certain threshold.

[0081] Figures 9 and 10 show sample pie charts. As described above, the display is extremely clear and can be visually interpreted in seconds. Furthermore, exact numerical values ​​are also displayed directly. The user can also select a pie chart (especially one that is noticeably asymmetrical) and examine the individual data / plots in more detail (for sample sections, see Figures 7 and 8 and Figure 11).

[0082] Figure 11 shows an example of a vertical movement of a right hip (from the "right-hand" perspective, i.e., the horse runs "from left to right" with its right side facing the observer). (The horse runs clockwise in a fictitious arena / ring, with the observer / camera in the center.)

[0083] The individual strides are shown, along with statistical averages. The individual stride phases (left hindquarters / right hindquarters) are clearly visible.

[0084] List of reference symbols

[0085] R10 Animal / Horse

[0086] R20 cameras

[0087] R30 synchronization module

[0088] R40 storage medium

[0089] R50 Master data, variable data

[0090] A10 Analysis Module (KL-supported)

[0091] A20 Correlation plausibility module

[0092] A40 training module (class-based)

[0093] A50 module for geometric correction

[0094] A70 Statistics Module

[0095] A90 output module

[0096] P1 - Pxx joint points and / or anchor points

Claims

Patent claims 1 . A computer-assisted method for detecting joint positions and / or angular movements of joints, in particular for analyzing gait mechanics, of living beings, in particular quadrupeds, in particular horses, ponies, camels and / or dogs, comprising: Providing at least one image, image stream and / or video stream showing a living being, in particular a quadruped, in particular from a lateral perspective, in particular running from a perspective lateral to a running direction of the living being, Determining at least one joint point and / or anchor point on the living being in the image, image stream and / or video stream by means of a machine learning model, in particular with the aid of a movement anatomy model of the living being, in particular a species- and / or genus-specific movement anatomy model.

2. The method of claim 1, further comprising: Determining a joint angle based on the determined joint points and / or using the machine learning model.

3. The method according to claim 1 or 2, wherein the image, the image stream and / or the video stream comprises: an image, an image stream and / or video stream for a first side of the living being, in particular wherein the living being runs from left to right, called first image, first image stream and / or first video stream, an image, an image stream and / or video stream for a second side of the living being, different from the first side, in particular wherein the living being runs from right to left, called second image, second image stream and / or second video stream.

4. The method according to claim 3, wherein a first machine learning model is used in determining the at least one articulation point and / or anchor point for the first image, the first image stream and / or the first video stream, and a second machine learning model is used for the second image, the second image stream and / or the second video stream.

5. The method of claim 3, further comprising Mirroring the first or second image, image stream and / or video stream about a vertical axis.

6. The method according to claim 5, wherein, when determining the at least one articulation point and / or anchor point, the mirrored and the respective non-mirrored image, image stream and / or video stream are processed by means of the machine learning model.

7. The method according to any one of the preceding claims, further comprising: Representing at least one joint position and / or joint angle in an output image, output image stream and / or output video stream, in particular based on the image, image stream and / or video stream and / or the movement anatomy model.

8. The method according to any one of the preceding claims, further comprising: Graphical representation of a maximum angle and / or a minimum angle of a joint point, in particular by displaying a joint movement amplitude.

9. The method according to any one of the preceding claims, further comprising: Graphical representation of at least one maximum angle and / or one minimum angle of a joint point in the form of a circle and / or pie chart.

10. The method according to claim 9, wherein in a right half of the circle and / or pie chart a right-sided maximum angle and / or minimum angle, in particular by Maximum angle and minimum angle formed angular ranges of a joint point of the living being are shown, and in a left half of the pie chart and / or pie chart a left-sided maximum angle and / or minimum angle, in particular angular ranges formed by maximum angle and minimum angle, of a joint point of the living being are shown.

11. The method according to any one of the preceding claims, wherein at least one universal timestamp is noted in one or more images, image streams and / or video streams, thereby enabling a meaningful combination of different images, image streams and / or video streams.

12. The method according to one of the preceding claims, wherein, in particular on the basis of images, image streams and / or video streams from angularly offset perspectives, at least one joint point and / or anchor point is determined in the form of three-dimensional point coordinates.

13. The method according to claim 12, wherein three-dimensional angular coordinates of joints and / or three-dimensional angular accelerations are calculated on the basis of the joint and / or anchor points and / or by means of the machine learning model.

14. The method according to one of the preceding claims, which is run through at least twice, wherein a first run is carried out on the basis of an image, image stream and / or video stream of a predefined quality, in particular high definition, wherein a second run, subsequent to the first run, is carried out on the basis of an image, image stream and / or video stream of a higher quality, in particular 4K.

15. The method according to claim 14, wherein the results of the first and second runs are compared with each other, in particular for quality assurance and / or error estimation.

16. The method according to one of the preceding claims, wherein a reference cuboid and / or reference rectangle is present in the image, image stream and / or video stream, in particular recorded together with the living being, in particular wherein the reference cuboid and / or reference rectangle is processed and taken into account by the machine learning model.

17. The method according to claim 17, wherein additional points with a predefined diameter and / or predefined distances from one another are used, in particular on and / or in the reference cuboid and / or reference rectangle.

18. The method according to any one of the preceding claims, wherein the joint and / or anchor points and / or the movement anatomy model further comprise a rider riding the living being.

19. The method according to any one of the preceding claims, further comprising: Determining lengths, distances traveled, speeds, and / or acceleration data.

20. The method according to any one of the preceding claims, Determination of a linear and / or angular acceleration at at least one joint and / or anchor point.

21. The method according to any one of the preceding claims, further comprising: Graphical representation of at least one maximum angular acceleration and / or one minimum angular acceleration of a joint and / or anchor point in the form of a circle and / or pie chart.

22. The method of claim 21, wherein in a right half of the pie chart and / or pie diagram, a right-sided maximum angular acceleration and / or minimum angular acceleration, in particular angular acceleration ranges formed by maximum angular acceleration and minimum angular acceleration, of a joint point of the living being are shown, and in a left half of the pie chart and / or pie diagram, a left-sided maximum angular acceleration and / or minimum angular acceleration, in particular angular acceleration ranges formed by maximum angular acceleration and minimum angular acceleration, of a joint point of the living being are shown.

23. The method according to any one of the preceding claims, further comprising: Automatic classification of a genus, species and / or breed of the living being by means of a class based on the image, image stream and / or video stream, in particular also recognition and / or classification of a rider who rides the living being.

24. The method of claim 23, further comprising: Providing and / or selecting an appropriate movement anatomy model based on the classification.

25. The method according to one of the preceding claims, wherein it is further ensured, in particular by a correlation plausibility module, that the joint points, anchor points, angles and / or variables calculated therefrom determined by the method are plausible, in particular plausible with respect to a genus, species and / or breed of the living being.

26. The method of claim 25, wherein implausible data is used in negative learning to further train and thereby refine the machine learning method.

27. The method according to one of the preceding claims, wherein furthermore, in particular by a statistics module, statistical anomalies and / or deviations are detected, in particular anomalies and / or Deviations beyond a predefined level of deviation with respect to other data on other living beings of the same genus, species and / or breed.

28. The method according to claim 27, wherein statistical anomalies and / or deviations in the context of negative learning are used to further train and thereby refine the machine learning method.

29. A method for training a machine learning model for use in a method according to one of the preceding claims, in particular comprising Providing an image, an image stream and / or video stream of a first side of a living being, in particular wherein the living being runs from left to right, called first image, first image stream and / or first video stream, as training data, Providing an image, an image stream and / or video stream of a second side of a living being, in particular wherein the living being runs from right to left, called second image, second image stream and / or second video stream, as training data, Mirroring the first or second image, image stream and / or video stream about a vertical axis.

30. The method of claim 29, further comprising Training the machine learning model through negative learning in the case of outputs of the machine learning model that statistically deviate beyond a predetermined level and / or are recognized as implausible, in particular with regard to data from other living beings of a corresponding genus, species and / or breed.

31. Use of the method for analyzing the motor characteristics of a living being, in particular deviations and / or asymmetries, in particular detecting, analyzing and / or quantifying lameness, in particular in horses.

32. A computer program comprising instructions which, when executed by a computer, cause the computer to effect the method according to any one of claims 1 to 31.

33. Computer-readable data comprising articulation points and / or anchor points which were generated using the method according to one of claims 1 to 31 or using the computer program according to claim 32, in particular also image and / or video data in which the said articulation points and / or anchor points are represented.

34. A computer-readable data carrier comprising the computer program according to claim 32 and / or the computer-readable data according to claim 33.

35. A computer, computer system or computer network which is configured to effect locally or non-locally the method according to any one of claims 1 to 31, the execution of the computer program according to claim 32 and / or the reproduction, processing and / or display of computer-readable data according to claim 33.

Images