Solution to fitness workouts

Literature Survey

In past years, several systems have been proposed for Fitness Workout. Different machines for Fitness Workout have been used. Below are some articles we have reviewed on the Fitness workout application.


READ FULL Literature Survey

In this study [21], the authors suggest a solution to static video surveillance systems that uses human activity recognition in real time. Using temporal pictures and convolutional neural networks, this method makes predictions about the actions of humans (CNN). CNN is a specific form of deep learning model that is able to learn features automatically by watching training films. Although the most advanced approaches have been proved to have a high level of accuracy, using them requires a significant amount of processing resources. Another issue is that the majority of approaches presume accurate prior knowledge of human situations. In addition, the majority of the existing techniques involve the creation of intricately handmade features for certain classifiers. Therefore, it is difficult to apply these kinds of procedures in situations that take place in the actual world. In this research, a unique CNN model that is constructed for real-time human action recognition is presented. The model is developed based on temporal pictures and a hierarchical action structure. The action layer, the motion layer, and the posture layer make up the three levels that are included in the hierarchical action structure. The most superficial activities are represented on the top layer, while position is shown on the lowest. The fact that there is one CNN in each layer indicates that there are a total of three CNNs at work in this model; the layers are merged in order to depict a wide variety of various forms of movement with a high degree of flexibility.

This chapter analyzes the classification accuracy attained by utilizing a three-dimensional convolutional neural network (3D CNN) model for video classification that was implemented in [22] earlier in this chapter. For the purpose of video classification, the 3D convolutional networks are preferable since, by their very nature, they perform convolutions in the 3D space. The information about this frame’s history in relation to the frames that came before it in time is referred to as temporal information. There are a few different approaches to video classification that have been developed, but each one has a few flaws that make it less than ideal. One example of such a technique is known as the convolutional neural network (CNN) model. It belongs to the category of deep learning neural network models that can be activated in an immediate fashion by the underdone inputs. In recent years, however, such models have been constrained to only accept inputs in a two-dimensional space.

In this study [23], a framework that combines render-based imagesynthesis and CNNs with the intention of addressing both of these challenges (Convolutional Neural Networks). Deep convolutional neural networks (CNNs) with a high capacity for learning can make good use of the ability that 3D models have to produce a huge number of images with a high degree of variety, as we believe that these models have the potential to generate. In order to accomplish this objective, we provide an image synthesis pipeline that is both scalable and resistant to overfitting, as well as an innovative CNN that has been specifically customized for the perspective estimation task. We demonstrate experimentally that the perspective estimation produced by our pipeline can significantly outperform state-of-the-art approaches when used to the PASCAL 3D+ benchmark. They create synthetic training images by superimposing images created from massive sets of 3D models on top of real-world photographs. CNN has employees who are trained to map photos to the ground truth of how objects are viewed. The training data consists of both genuine photos and ones that were artificially generated. To determine the perspectives from which objects in real photos are seen, the learnt CNN is put to use.

In many fields of medical and laboratory research, as well as clinical practice, imaging has become indispensable. Radiologists identify and quantify tumors from MRI and CT scans, and neuroscientists detect regional metabolic brain activity from PET and functional MRI scans. Biologists study cells and generate 3D confocal microscopy data sets; virologists generate 3D reconstructions of viruses from micrographs; radiologists identify and quantify tumors from MRI and CT scans, and neuroscientists detect regional metabolic brain activity from PET and functional MRI scans. Computerized quantification and visualization tools are required for the analysis of these various image types. Until recently, only expensive UNIX workstations and customized software could be used to perform 3D visualization and quantitative analysis of images. Much of today’s visualization and analysis can be done on a low-cost desktop computer with the necessary graphics hardware and software. This paper introduces [1] a general-purpose image processing and visualization program that is extensible, platform-independent and specifically designed to meet the needs of an Internet-connected medical research community. MIPAV (Medical Image Processing, Analysis, and Visualization) is an application that allows for clinical and quantitative analysis of medical images over the Internet. Researchers and clinicians at remote sites can easily share research data and analyses using MIPAV’s standard user interface and analysis tools, enhancing their ability to study, diagnose, monitor, and treat medical disorders.

         On the basis of this[2], this study employs the physical traits of personal trainers as image target detection indicators. This study uses infrared capture as the basis of image capture detection technology, uses FCM clustering as the fuzzy image background segmentation algorithm, uses k-means clustering analysis to study the gray histogram, and proposes a composite classification feature tracking method for trainer image tracking in accordance with the theory that the human body will dissipate more heat during the fitness process. When combined with the experimental research, the study demonstrates how the research method improves the detection rate of the human target by making use of the benefits of the composite classification feature. As a result, it is a highly efficient and real-time algorithm for detecting humans in infrared images.

A machine learning approach is suggested in this paper[3] to evaluate user activity on a platform for at-home workouts. Keep is an at-home workout program that offers a variety of exercise options, including yoga, cycling, running, and fitness diet advice. They compared the four supervised learning algorithms support vector machine, k-nearest neighbor, random forest, and logistic regression using the combined training set data of 7734 Keep users. The receiver operating curve analysis showed that random forest had a higher overall discrimination verification power than the other three models. 850 test samples were classified using the random forest model, and an accuracy rate of 88 percent was attained. This method can forecast users’ continued use of the home workout application after installation. On Keep, they took into account 18 variables that were predicted to have an impact on the determination of continuous participation. Keep certification was the main factor influencing the study’s findings.

The goal of this study[4] is to create a deep learning pipeline to identify signals of adverse events (DS AEs) connected to dietary supplements on Twitter. Materials and procedures From 2012 to 2018, 247 807 tweets mentioning both DS and AE were collected by us. They created a specific annotation guideline for DS AEs and annotated 2000 tweets with biomedical entities and relations. With a CRF classifier, we optimized and compared the performance of the BioClinical-BERT, PubMedBERT, ELECTRA, Roberta, and DeBERTa models for the concept extraction task.

By considering these factors we recommend a Mobile-application as a solution and hope it will be of help to Fitness Workout who are not satisfied with their Fitness performance levels and who do not have the facilities to obtain the best and free solutions to workout or the time, can access our mobile application and get the help they need.

Research Problem

Lack of understanding of the workout machine by users Using the exercise equipment improperly results in severe health consequences. problems in the future Most gyms don’t have skilled gym trainers. Employing trainers is expensive. The current application is essential for conclusively resolving the research challenge.

The majority of gym-goers utilize supplements without proper knowledge. Regular usage of the product without adequate information can create major health problems. It is difficult to obtain exact replacement products from other brands. having difficulty finding inexpensive alternatives. As a solution, the system facilitates the systematic resolution of research problems by its users.

Previously, image processing was the primary focus. Users wish to understand more about video processing at present. Researchers seek to bring the same computing capacity to video processing that CNN brought to image processing. It will assess CNN’s video processing capabilities. The capabilities of the various CNN models available for video processing will be discussed.

The majority of supplement consumers receive their supplements without adequate knowledge, which can result in adverse effects. Taking a supplement without sufficient understanding is too dangerous. The current application is required to solve the research problem in its entirety.

Overall there is no combined app with these components, this may a unique mobile app to over come these problem.

Research Objectives

The Proposed system will be carried out
under four main components.

  • Sub Objective 1: Product analysing using machine learning which would be able to provide the details of the supplement to maintain a healthy body.
  • Sub Objective 2: Machine trainer using image processing with a neural network which will assist as a virtual trainer.
  • Sub Objective 3: Activity analyser using machine learning that could provide the amount of the activity and correct it as possible.
  • Sub Objective 4: Product adverse notifier using text analysis, which can aid other users to find supplement side effects.


Figure 1 shows a block diagram of the Integrated Gym system: A Supporting application for Enhancing the fitness workout. The application’s primary goal is to enable support for Gym as a helping tool. The mobile application will aid in the fitness workout and assist users who are struggling to get a product analyser, gym assistant, or Activity analyser for the correct healthy fitness, and Product adverse.

The Proposed system will be carried out under four components.

  1. Product analysing using machine learning which would be able to provide the details of the supplement to maintain a healthy body.
  2. Machine trainer using image processing with a neural network which will assist as a virtual trainer.
  3. Activity analyser using machine learning that could provide the amount of the activity and correct it as possible.
  4. Product adverse notifier using text analysis, which can aid other users to find supplement side effects.
  5. Give the solution to fitness workouts using a mobile app.


Different types of CNN algorithms and models will be used to examine the data. After the ML (Machine Learning) model has been trained, the binary classifier results will be categorized. Python was chosen as the programming language, along with the libraries, Deep Learning, ML framework, K-means clustering Model to Implementation, and Image processing for finding the product and Video processing.

The Machine Learning component of this research will use a convolutional neural network to analyze the images captured via the video device. Once the ML (Machine learning) model has been trained, it will categorize the binary classification results. Python was selected as the programming language, along with the following libraries: NumPy for handling array data of images, OpenCV for image classification, TensorFlow, and Keras to handle the ML (Machine learning) framework.



  • Topic Assessment

    The initial step where we got the feedback and comments from the panel.

  • Proposal Presentation

    The first initial idea pitching to the panel with aimed methodologies and technologies to be used in the making of The Best-You: Gym Based Machine Learning Application.

  • PP1

    This was where the 50% progress were presented to the penal.

  • PP2

    This was where the 90% progress were presented to the penal.

  • Final Presentation

    The final launch of The Best-You: Gym Based Machine Learning Application kick started from this point, where we presented 100% completed final product to the final.

  • Research paper

    The biggest achievement in the academic career was achieved through this phase in research. By submitting the research paper to the ICAC conference.


Proposal Document


Final Document




Status Document



Presentation Final


Proposal Presentation


PP1 Presentation


PP2 Presentation


About Us

Ms. Hansi De Silva

Ms. Hansi De Silva

(Main Supervisor)
Ms. Geethanjali Wimalaratne

Ms. Geethanjali Wimalaratne



Sasfak Ahamed AG.

Sasfak Ahamed AG.

Ahamed MBB.

Ahamed MBB.

Hasni MNN.

Hasni MNN.

Shajith MSM.

Shajith MSM.