DE102020100366A1 - Procedure for 3D motion analysis and instant feedback for training exercises - Google Patents

Abstract

The invention relates to a method for 3D movement analysis and immediate feedback for training exercises, whereby a person is captured by means of a 3D sensor, raw data in the form of image data and depth data of the person are captured by means of the 3D sensor and this data is passed on to a processing unit and in it are processed, with articulation points being extracted from the raw data and a division into elementary exercises and their temporal normalization for a subsequent movement analysis.

Description

The invention relates to a method for 3D movement analysis and instant feedback for training exercises and is used in particular for the field of medical training therapy and in the fitness market, for prevention, in home training and in competitive sports.

In medical training therapy (MTT) it is common for patients to complete a program to restore strength and mobility after a previous operation (e.g. implantation of an artificial hip joint). This is done on an inpatient or outpatient basis under the guidance of a therapist. The current personnel and financial situation in the healthcare system makes it necessary for a therapist to care for up to 15 patients at the same time. As a result, the therapist cannot guide the patient professionally at all times. As a consequence, an incorrectly performed exercise can make the training therapy less effective or even cause a harmful effect. It is essential that the training exercises are carried out as precisely as possible. To prevent this, technical procedures to support the therapist must be developed. These must be able to record the movement in three-dimensional space and to couple it back to the patient in such a way that the quality of movement is improved. Such a solution would result in increased strength and mobility, which significantly increases the quality of life of those affected.

Previously commercially available training devices use measuring sensors to record the number of sentences and repetitions as well as the amplitude and speed of movement. Movement corrections are not conveyed to the patient. Feedback-based training is often used as part of rehabilitation, especially in balance and gait training for stroke patients. Various studies have already shown that the use of feedback training by means of immediate information in rehabilitation has proven itself ( Banala, Kim, Agrawal & Scholz, 2009; Hirokawa & Matsumura, 1989; Kim, Ogilvie, Shimabukuro, Stewart & Shin, 2015; Raschner, Lembert, Mildner, Platzer & Patterson, 2008 ). In the MTT of orthopedic patients, however, real-time feedback has hardly been used in practice. In Lam, HajYasien & Kulic (2014) inertial sensors were used. Using an elaborate calculation process, the movement data must first be converted into angles before the test person receives the feedback. In addition, the inertial sensors used must be attached to the test subject. In addition, some systems use so-called markers (Kim et al. 2015). However, this system has no error display, only a movement template that shows the stride length.

Detection of explicit error patterns has not been possible until now. An error pattern is a movement that does not correspond to the target movement and has a negative influence on the quality of the exercise performed. During hip abduction on the cable pull, many patients tend to bend their upper body instead of spreading the leg. This makes the exercise easier for the patient, but does not achieve the desired effect of strengthening the hip muscles. The state of the art describes ( Huang, Cheng & Chiang, 2013; Khan, Lin, Guan &Guo; Su, Chiang & Huang, 2014 ) a system based on dynamic time warping, but with these only the similarity of the executed movements with reference to a target movement is determined.

There are currently no systems on the market that convey specific error patterns to the patient for rehabilitation exercises. Rather, systems can be found in the fitness sport sector that measure amplitudes and repetitions using force, acceleration and distance sensors built into devices and pass them on to the user. There are only devices that recognize error patterns during the execution of the exercise using 3D skeleton data based on manually established rules. These rules are mostly based on body geometries, such as angles or heights, which are measured on the skeleton.

A company founded by sports scientists and engineers advertises digitalized strength training. An inertial sensor, which can detect the movement in a 3D coordinate system, is attached to the device (for example a dumbbell). The documentation of the parameters speed, distance and power as well as the execution of the exercise, which is output by the dumbbell locus, are used for training control.

In the pamphlet DE 103 31 110 A1 a method for simulating musculoskeletal stress on a patient is described. This method should be suitable for monitoring surgical interventions as well as for monitoring rehabilitation measures. Individual musculoskeletal parameters and individual movement parameters are determined. From these, individual musculoskeletal loads are automatically determined. These are compared with the target parameters. If this is not achieved, the corresponding parameters are corrected. The comparison is made computer-assisted. The determination of the individual musculoskeletal parameters is based on the position of the joints. A The parameters are set using the musculoskeletal parameters. The position and / or the orientation of the joints play a central role. The data determined in this way are saved and compared with reference data from a database. This allows optimal individual musculoskeletal stress data to be derived. Discrete musculoskeletal parameters are stored in the database, i.e. parameters that can be assigned to individual joints. The calculation of the individual musculoskeletal load is carried out with the aid of biomechanical and / or mathematical models. Furthermore, it is stated that the models are adapted using individual musculoskeletal parameters. The results of the process can be displayed graphically or numerically.

The method can be used to carry out and monitor rehabilitation processes. The determination of the individual parameters, by manual measurement or by e.g. computed tomography or sensor technology, can take place automatically. In the method shown, individual gait parameters can also be implemented. No normalization of the musculoskeletal model is created. This means that there is no independence from body size. There is no time normalization. This includes the effect of the training movement carried out at different speeds. There is no visual real-time feedback that represents an avatar and quickly points out incorrect exercises.

From the pamphlet DE 600 23 463 T2 a device for exercising posture is known. This is designed so that it is placed on the floor and has a movable seat. There is also a fixed backrest and a gymnastics element is implemented. A receiving means is installed on the seat. The seat and the recording means are equipped with sensors that detect the exact posture of the exerciser. During the exercise, the person's posture is recorded. Incorrect postures are recognized and signaled optically or acoustically. The device is also equipped with a metronome so that the exercise processes can be monitored over time.

There is only a mechanical measurement with sensors. There is no contactless measurement. This means that only the movement of the device is recorded, but not whether the person is performing the movement correctly. There is no real-time visual feedback. No 3D skeleton model is determined. No specific error patterns are recognized.

The pamphlet DE 10 2016 006 009 B4 discloses a method of exercise monitoring. In this case, at least two sensors are arranged on the exerciser or one sensor on the person and one on the exercise device. These sensors detect the movements and transmit them to a processing unit. This determines an actual movement pattern from the data. A target movement pattern is transferred from a database and both are compared. If this comparison deviates, an optical or acoustic error message is issued. The data processing takes place in real time. Thus, the trainee can react immediately to any mistakes during the exercise. The sensors are installed either on the device or on the person. There is no contactless measurement. There is no real-time visual feedback. No 3D skeleton model is determined. No specific error patterns are recognized.

In the pamphlet DE 10 2017 116 558 A1 describes a method for performing movement sequences and a training device for guiding movement sequences. The device consists of a reflective surface in the form of a mirror and a sensor which is directed at the exerciser. The sensor detects the viewing direction of the eyes with respect to the reflection surface. In addition, the position of a representative body part of the person is recognized. The system generates a target area, where the body part is located or should move, and displays it on the reflection surface. The target area can be displayed as a silhouette of the exerciser. The impression is given as if it were a mirror image. The control unit can have access to a database in which the exercise sequences are stored in the form of target positions. An actual silhouette is generated by the sensor signal or signals and is also displayed on the mirror surface. If the exercise is carried out by the exerciser as the target specification provides, both silhouettes appear roughly congruent. No 3D skeleton model is determined. No specific error images are recognized, only whether the position in 2D (projection surface) is correct.

However, the prior art has some disadvantages. Previous systems are not able to explicitly record specific error patterns and also work on a rule-based basis and are not easily adaptable to new training exercises. In addition, there is no visual real-time feedback using avatars, which shows error images. Motion detection in 3D is also insufficient. Systems that work marker-based are quite precise, but not practical, as it takes a lot of time to apply the markers, which means that it is not possible to quickly switch between patients.

Exact execution of the exercises is crucial for orthopedic rehabilitation, as otherwise the patient may be injured. Since marker-based systems are ruled out because of the effort involved, it is necessary to record the patient's movement without any additional preparation so precisely that a movement analysis is possible. For this, a body model of the person with the most important bones and joints must be created.
The determination of explicit error patterns is necessary, as this is the only way for the patient to receive an explicit correction note via feedback. In the case of the above-described tilting of the upper body during hip abduction, the request must come not to tilt it.
Rule-based procedures have to be configured manually for each individual exercise and require detailed technical knowledge. A therapist is unable to develop his own exercises.
The inadequate feedback from existing systems means that possible errors are not adequately passed on to the user. This means that the movement cannot be corrected. Previous systems only count the number of repetitions and recognize whether a movement was carried out incorrectly, but not what is specifically wrong.

The object of the invention is to develop a method for 3D movement analysis and instant feedback for training exercises, which can precisely record the movements of a person, automatically derive error images from them and output real-time visual feedback to the person.

This object is achieved with the characterizing features of the first claim.

Advantageous refinements result from the subclaims.

The method is used for 3D movement analysis and immediate feedback for training exercises, whereby a person is recorded by means of a 3D sensor in such a way that raw data in the form of image data and depth data of the person are recorded by means of the 3D sensor, these data being forwarded to a processing unit and be processed in this. Articulation points are extracted from the raw data. According to the procedure, there is a division into elementary exercises and their temporal normalization for a subsequent movement analysis.

Error images are preferably recognized and displayed to the person via a visualization unit. Mistakes include wrong movements or postures as well as too fast or slow execution of the exercise. This means that the method also records the time for a single movement of the exercise.

In an advantageous embodiment, these articulation points or the anatomy of the person are normalized.

The starting point for all calculations is the optical 3D sensor, especially in the form of a camera. It is ensured that the person whose movement is to be analyzed is placed completely within the field of view of the 3D sensor. Depending on the opening angle of the sensor, there is a different distance between the sensor and the person. The output of the sensor consists of an RGB image and the corresponding depth coordinates of the corresponding pixels.

The extraction of the points of articulation is preferably implemented as described below. The person is detected by means of the sensor, and their 3D position in space is recorded. The determination of the individual articulation points, which are required for motion detection, is derived from this. Neural convolution networks are used to detect every main joint point of the human body, so that a total of 19 joint points are extracted. In order to obtain a 3D position from these points in a coordinate system with an x, y and z axis in relation to the camera position, the points are projected back onto the corresponding depth data. Based on the information of the horizontal and vertical resolution in pixels, the horizontal and vertical field of view and the 2D position from the neural convolution network, the 3D position is determined to the centimeter. The result is called a skeleton model. The method mentioned is only to be seen as an example, although other methods can also be used.

The resulting skeleton model must then be standardized in order to achieve independence from the sensor or the anatomy of the test subject. This corresponds to a generalization for later machine learning. First, the sensor coordinate system / camera coordinate system, with the 3D sensor representing the origin of the coordinate system, is converted into a local, person-dependent coordinate system. The hip of the captured person represents the new origin. The other points of articulation are calculated relative to this. In a second step, the local coordinate system is converted into a hierarchical system of coordinates. A point of articulation only depends on its next following point of articulation. For example, the left foot is derived from the left knee, which is derived from the left hip. That is, a rotation matrix describes the relationships between two consecutive points of articulation. In a final step, the distance between the individual points of articulation is normalized to one. This means that the final model is independent of the size of the person and general parameters can be set up.

For the evaluation of individual exercises, these are divided into elementary exercises. This is also known as repetition. An elementary exercise describes exactly one execution of this exercise, for example exactly one squat. Since the movements occur cyclically, this can be modeled with a sinusoidal curve for individual joint points. To make a split, the angle between two interacting hinge points is used. For hip abduction, for example, the angle between the left and right ankle spanned across the hip is taken as the basis. The angle reaches a minimum when the feet are together and a maximum when the legs are apart. This local minimum is now detected and defined as the start or end of an elementary exercise. The points between two minima represent the actual elementary exercises. In addition, successive minima can be used to determine how quickly an exercise was carried out and whether it is within the target range.

After the division into elementary exercises, the time standardization of the elementary exercises takes place. The time factor must be normalized for motion detection so that it can be compared with a previously recorded reference. This ensures that the speed of execution does not play a role in the assessment of the quality of the exercise or the detection of error images. Dynamic Time Warping (DTW) is generally suitable for this task. However, since this procedure requires that the movement has been completed, Incremental Dynamic Time Warping (IDTW) can be used as an alternative. A cumulative cost matrix K is calculated for each individual joint point, while the DTW values in K are the cumulative Euclidean differences between a reference and a current coordinate. Every time a new image is taken and new points of articulation of the skeleton model are calculated, a new column is appended to K. The image from the reference recording that best fits the last recorded image is determined by determining the minimum DTW value in this column. In this way, the part of the reference that best fits the current partial sequence is selected.

In addition to the procedure mentioned, this approach is expanded to the effect that the distance between the reference and the current image is calculated in each image for each articulation point. This corresponds to the distance vector D. As a result, it is not the minimum DTW value that is selected, but the respective distance vectors. In order to reset the cost matrix K after each repetition, the previous recognition of the elementary exercises is used. The cost matrix K is reset when the end of a repetition has been recognized, that is, all elements are deleted.

According to the method, motion is detected on the basis of error images. In a first step of the movement analysis, a feature vector M is set up which contains a subset of all distance vectors. The decisive factor is which exercise and which joint point is involved. This feature vector represents the input data for the machine learning process. This version is a support vector machine (SVM). In particular, it is a multi-class SVM with a one-vs-all approach. This SVM was trained with many feature vectors M in advance. It is known in which sequence error patterns occur. These have been annotated accordingly. If there are x error patterns, x + 1 classes exist, with the additional class standing for correct movement execution. It should be noted that the classification is ordered hierarchically. This means that only one class can occur at a time. If none of the classes is found, correct movement can be assumed.

The communication between the 3D sensor and the visualization unit takes place via a standard TCP / IP protocol. Data about the person, the training exercise and the course of the exercise currently being carried out are transmitted. A transmission takes place on a frame basis. This guarantees real-time capability for the visualization.

In an advantageous embodiment, the method is linked to a database, with all information or data relating to the person or persons as well as the training plans being stored in the database. It specifies exactly how many sets and repetitions are to be completed in a single training exercise. In addition, the skeleton model information from all articulation points as well as the detected error images are saved for each image.

In order to motivate the test persons to carry out the therapy exercises and to guarantee that the exercise is carried out as precisely as possible, real-time visual feedback is implemented. The system consists of a visualizing part and a part for data management.
For the personal database entries explained in the above section, there is a graphical user interface for data entry available. The therapist can enter the data for the individual training scenarios individually for each patient. The patient can start the training exercise independently.
The articulation points previously described in the skeleton model are connected to form a three-dimensional wire frame model. This is referred to as the avatar in the following. The avatar is animated in real time. The avatar can be designed individually by the user. If error images are recognized beforehand by the movement analysis, the affected parts of the body are colored using realistic traffic light colors (red, yellow, green). Red and yellow stand for errors and green for correct execution. In addition, instructions are given on how to correct the error, for example in the form of a request to “keep your upper body straight”. Corrections in posture can then be made immediately. The individual steps of an exercise are counted and displayed in real time. After the repetitions necessary for the exercise have taken place, a statistical evaluation is visualized for the user. These statistics give a precise overview of the training progress and any problems. In addition, the amplitude angle during execution and the accumulated individual errors are recorded and visualized.

• • Distanz zwischen den Fußgelenken ist kleiner als 250 mm.
• • Der Flexionswinkel des linken Knies ist größer als 160 Grad.
• • Der Flexionswinkel des rechten Knies ist größer als 160 Grad.
• • Der Winkel zwischen linker Schulter und linkem Fußgelenk mit der linken Hüfte als Basis ist größer als 150 Grad.

In order to be able to record an exercise, it is relevant when a person is ready for it. A previously defined posture is detected as a start event for the analysis. Such a posture can look like this, for example. To trigger the event, a person must stand up straight, straighten their knees, and keep their feet close together. The parameters are as follows:

• • The distance between the ankles is less than 250 mm.
• • The flexion angle of the left knee is greater than 160 degrees.
• • The flexion angle of the right knee is greater than 160 degrees.
• • The angle between the left shoulder and the left ankle with the left hip as the base is greater than 150 degrees.

If these four conditions are met for more than 4 seconds, the start event is detected in an exemplary embodiment. Alternatively or additionally, the start event can be triggered by a body movement.

According to the method, post-filtering of the raw data is preferably carried out, which guarantees temporal smoothing and leads to greater robustness. The task of post-filtering is to exclude short temporal deviations from the standard behavior. In particular, noise in the depth data can lead to a change in the 3D position of an articulated connection, which also affects the method. In addition, a person can be on the boundary of the classes. Temporal smoothing is necessary to avoid jittering or image instability. This is implemented using a mean value filter that is applied to the time series of the previously created error labels. Preferably, a medium size 3 or 5 filter is sufficient to remove noise and jitter.

In an advantageous embodiment, the progress of the person is documented by means of storing the personal real-time data, whereby the storing of the personal real-time data can be viewed at any time. In the event that the person is a patient, it is possible for the therapist to create a statistical evaluation of the patient's progress and to adapt and improve the training accordingly. A targeted review of an individual training session can also take place together with the test person. For this purpose, recorded movements can be extracted from the database and then visualized. It is possible to adjust the speed of execution of the movements or to pause them. This enables the therapist to provide the patient with detailed assistance, which is far more targeted and goes beyond the previous movement rehabilitation. In addition, the progress documentation can be relevant for insurance providers. The method can be used for training with your own body weight or for equipment training.

• • Die Person muss mit einem 3D Sensor erfasst werden.
• • Es ist eine exakte Bestimmung des Skelettmodells notwendig.
• • Es müssen Normierungen der Bewegungen stattfinden, damit die Anatomie der Person und die Geschwindigkeit der Bewegungsausführung keinen Einfluss mehr haben.
• • Es muss eine Klassifizierung der Fehlerbilder mittels Methoden des maschinellen Lernens stattfinden. Dabei beruhen die Modellinformationen auf Beispieldaten und nicht auf expliziten Regeln.
• • Es muss ein visuelles Echtzeitfeedback realisiert werden. In diesem muss ein Avatar sichtbar sein, welcher der Bewegung präzise folgt. Fehler müssen anatomisch korrekt angezeigt und Korrekturvorschläge gemacht werden. Es kann eine Statistik zu häufigen Fehlerquellen erfolgen. Zudem werden die Übungen zeitlich und räumlich quantifiziert.
• • Eine Datenback muss implementiert werden, in der die spezifischen Informationen des Patienten beinhaltet sind. Diese garantiert eine individuelle Behandlung.
• • Es muss eine Schnittstelle zwischen 3D-Bewegungsanalyse und visuellem Echtzeitfeedback geschaffen werden.

The following points must be completed in order to carry out the procedure described.

• • The person must be recorded with a 3D sensor.
• • An exact determination of the skeleton model is necessary.
• • The movements must be normalized so that the anatomy of the person and the speed at which the movement is performed no longer have any influence.
• • The error patterns must be classified using machine learning methods. The model information is based on sample data and not on explicit rules.
• • Visual real-time feedback must be implemented. In this an avatar must be visible, which precisely follows the movement. error must be displayed in an anatomically correct manner and suggestions for correction made. Statistics can be provided on frequent sources of error. In addition, the exercises are quantified in terms of time and space.
• • A data back-up must be implemented, which contains the specific information of the patient. This guarantees an individual treatment.
• • An interface between 3D motion analysis and visual real-time feedback must be created.

The invention is explained in more detail below using an exemplary embodiment and associated drawings.

• 1 einen Aufbau für das erfindungsgemäße Verfahren,
• 2 die Verfahrensschritte für die Bewegungsanalyse,
• 3 ein verfahrensgemäßes Skelettmodell.

Show it:

• 1 a structure for the method according to the invention,
• 2 the procedural steps for the motion analysis,
• 3 a procedural skeleton model.

1 shows a structure for the method according to the invention. An optical 3D sensor 1 takes the RGB image as well as the depth data of a person P. and forwards the information to a processing unit 2 further. In this, a movement analysis including the determination of defect images takes place. The processing unit 2 is bidirectionally linked to the visual real-time feedback. This also includes the database and the automatic progress documentation. As a visualization unit 3 a display is provided.
Within the processing unit 2 will be according to 2 implemented the following steps to determine the error patterns. The joint points are extracted on a frame-based basis from the raw data of the optical 3D sensor. These hinge points are then normalized. A division into elementary exercises and their temporal normalization is the basis for the subsequent movement analysis, whereby specific error patterns are recognized. Post-filtering guarantees temporal smoothing and leads to greater robustness.
The 3D sensor 1 , the processing unit 2 and the visualization unit 3 can be separate systems or combined in a single system. For example, a display with a 3D sensor and a computer unit can be combined in one housing.

In 3 is a generated skeleton model 4th shown that from individual joints 5 is composed. If fault patterns are recognized by the motion analysis, the affected body parts are 6th colored by realistic traffic light colors (red, yellow, green). Red and yellow stand for errors and green for correct execution.

• • Das verwendete Skelettmodell zur Extraktion der Skelettpunkte ist wesentlich genauer als bisherige Ansätze.
• • Die Normierung des Skelettmodells gewährleistet eine Unabhängigkeit von individuellen Körperproportionen.
• • Die zeitliche Normierung mit der IDTW gewährleistet eine Unabhängigkeit von der Ausführungsgeschwindigkeit einer Übung.
• • Die Bewegungsklassifikation zur Fehlerbilderkennung beruht auf Methoden des maschinellen Lernens und ist nicht regelbasiert. Dies resultiert in einer erhöhten Genauigkeit und einer Zeiteinsparung.
• • Ein visuelles Echtzeitfeedback mit Avatar in Ampelfarben sowie konkreten Korrekturhinweisen. Zudem erfolgt eine statistische Erfassung der Übungsausführung. Es existiert eine einfache und verständliche Bedienbarkeit.
• • Eine Verlaufsdokumentation mittels abgespeicherten Gelenkpunkten bietet enorme Vorteile für Patienten, Therapeuten und Kostenträger.

The method according to the invention has the following advantages over the prior art:

• • The skeleton model used to extract the skeletal points is much more accurate than previous approaches.
• • The normalization of the skeleton model ensures independence from individual body proportions.
• • The time standardization with the IDTW ensures independence from the execution speed of an exercise.
• • The movement classification for fault pattern recognition is based on machine learning methods and is not rule-based. This results in increased accuracy and time savings.
• • A visual real-time feedback with avatar in traffic light colors as well as specific correction instructions. In addition, there is a statistical recording of the execution of the exercise. There is a simple and understandable usability.
• • Documentation of the progress by means of stored articulation points offers enormous advantages for patients, therapists and payers.

List of reference symbols

1

Optical 3D sensor

2

Processing unit

3

Visualization unit

4th

Skeleton model

5

joint

6th

Body part

P.

person

QUOTES INCLUDED IN THE DESCRIPTION

This list of the documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• DE 10331110 A1 [0007]
• DE 60023463 T2 [0009]
• DE 102016006009 B4 [0011]
• DE 102017116558 A1 [0012]

Non-patent literature cited

• Banala, Kim, Agrawal & Scholz, 2009; Hirokawa & Matsumura, 1989; Kim, Ogilvie, Shimabukuro, Stewart & Shin, 2015; Raschner, Lembert, Mildner, Platzer & Patterson, 2008 [0003]
• Huang, Cheng & Chiang, 2013; Khan, Lin, Guan &Guo; Su, Chiang & Huang, 2014 [0004]

Claims (12)

Method for 3D movement analysis and immediate feedback for training exercises, whereby a person is recorded by means of a 3D sensor, raw data in the form of image data and depth data of the person are recorded by means of the 3D sensor and this data is passed on to a processing unit and processed therein, Articulation points are extracted from the raw data and a division into elementary exercises and their temporal normalization takes place for a subsequent movement analysis. Procedure according to Claim 1 , characterized in that error images are recognized and displayed to the person via a visualization unit. Procedure according to Claim 1 or 2 , characterized in that the hinge points are normalized. Method according to one of the Claims 1 to 3 , characterized in that the articulation points are extracted based on frames. Method according to one of the Claims 1 to 4th , characterized in that a skeleton model is generated from the hinge points. Method according to one of the Claims 1 to 5 , characterized in that the visual feedback is real-time capable. Method according to one of the Claims 1 to 6th , characterized in that the distance between a reference value for the associated exercise and the current image is calculated on a frame-based basis for each joint point. Method according to one of the Claims 1 to 7th , characterized in that a previously defined posture and / or a body movement is detected as a reference for the analysis. Method according to one of the Claims 1 to 8th , characterized in that the exercise sequence can be saved and retrieved. Procedure according to Claim 9 , characterized in that a progress documentation is carried out by storing the personal real-time data, with a statistical evaluation of the personal progress being created. Method according to one of the Claims 1 to 10 , characterized in that movement sequences of the articulation points are smoothed over time via post-filtering. Method according to one of the Claims 1 to 11 , characterized in that the articulation points of the skeleton model are connected to a three-dimensional wire frame model and that the animation of the wire frame model takes place in real time.

Images