PureSwing.ai

I have started a golf swing analyzer with a friend who attends a different school so for our project I will continue to work on this. Our product will be able to take in a video of a golfers swing and, using computer vision and machine learning, can detect faults in their form. The app will then provide feedback to user and drills to improve.

Golfers need on demand swing analysis while on the golf course to improve their swing mechanics

An app which can take in a video of your swing and can provide feedback, tips, and drills to correct any problems.

Our goal with implementation is to first create an MVP which simply tracks the golf club. This can provide golfers assistance in determining if their swing path is off-plane. Creating this MVP will allow us to receive feedback from customers and determine if we should continue developing features to turn PureSwing.ai to a full fledged golf swing analysis platform or to pivot to a different objective. Through the MVP, we want to ensure that we are building something that people actually want and are willing to pay for. So far we have gathered just over 700 videos of golf swings and have had three of them annotated (consisting of just over 1000 frames). We are currently in the process of developing our neural network architecture to find the location of a golf club in any given frame. We have also created a simple app for creating training data for our specific use case (golf swings recorded from a phone). We hope to have an MVP on the market by the end of the month.

The KT decision matrix and KT decision matrix have been completed and are attached below. Along with this, the ECE design site has been bookmarked and the expense reports have been reviewed.

The cost of failure is not too bad other than a bad review if the neural network provided incorrect guidance for a user's swing. To combat this, we plan to develop strict protocols that the network must abide by to give guidance. If the network is not adequately confident that a certain thing is wrong, the guidance will not be displayed. This may result in no feedback on some of the video uploads but we believe this is better than providing incorrect feedback. To improve and hopefully prevent failures of the network, we plan on collecting the data from all the uploads, get the videos annotated, and use them as the training data for the network. This will allow the network to be trained on data that it will actually be received when in production which will improve accuracy. Using continuous integration, users will always be using the most up-to-date network we have.

Since PureSwing.ai is a software product and has no physical aspects, disposal after the useful life has ended is not an issue. Where we do have to look at the ethics and liability is regarding the misuses and changes to the product during its lifetime. The neural network will be trained and improved as we have access to more data and financial resources. To keep up with these changes we will be implementing continuous integration methods so we are deploying updates continuously. For foreseeable misuses, if the algorithm recommends the incorrect guidance to the user, it could leave us with a less than satisfactory review but it is not life threatening. We will be sure to stay on the conservative side when creating recommendations.

Attached below

Through this weeks exercise we were able to create a system model which describes what the experience from the user side and a process model which describes all the magic that occurs on the backend when a user uploads a video of their swing. The models can be found below.

After searches on USPTO for the name "PureSwing.ai" the closest thing I found was Pure Swing Products which were training aids. I do not think this conflicts with our desired name and branding. A screenshot of the search result can be found below. I also created a quick logo for the app which can be seen at the top of the page.

Final Solution: Create mobile app to automatically analyze golf swing videos

1. Get joints of user in video

   • This will be done using wrnch.ai

2. Get golf club coordinates

   • Needed to get swing path, swing speed, and tempo

3. Map joints on 3d graph to monitor motion

   • Get json data produced by wrnch.ai

   • Create 3d graph animation using matplotlib

   • View results and monitor for any unexpected results

4. Implement known best golf swing technique checks

   • Example: Ensure shoulders and hips turn at same time. if not, notify the user

5. Gather data to train model

   • Upload professional golfer swings to wrnch.ai

   • Store data for each swing

   • Provide data to machine learning model to find important aspects of each swing

6. Test model on bad golf swings

   • See what the model notices on bad golf swings

   • Continue to improve

7. Create app to easily interface with software on mobile

   • make platform for golfers to upload their swings and receive feedback in a very easy to use way

Source/Cause:

Golfers do not perform proper mechanics when swinging -> Why? -> They never received proper instruction -> Why? -> Golf lessons are expensive, do not align with their schedules, and are not available when they need them

Kepner-Tregoe (KT) situation analysis:

• What?

  • We know golfers want to improve their swing to perform better

  • We do not know if golfers are willing to receive advice from a computer rather than an instructor

  • The goal is to provide real time tips to golfers who upload their swing

• When?

  • The problem occurs when the golfer continues to perform poorly either on the golf course or at the driving range

• Who?

  • Professional golfers will create the data for the AI/Machine Learning model to learn

  • Beginner to intermediate golfers will use the model to find differences between their swing and a golf professional's swing

• Where?

  • The problem occurs at the driving range or on the golf course

• Why?

  • The problem is important to any golfer who wants to improve their swing when they want to

• How?

  • The golfer can get immediate feedback on their swing by uploading a video of their swing to the app

• Objective (why?) : Allow golfers to get on demand feedback on their golf swing.

• Background (who? where?) Golfers can use the product on the golf course or at the driving range through their mobile device.

• Methodology (how? when?) Golfers will upload a video of their swing and, utilizing AI and computer vision, will receive feedback

• Expected results (what?) The feedback expected will be notifying what part of their swings needs work and recommend drills to fix their problems.

• Costs (how much will it cost?)

  1. 1. 1. User costs: $10/month

        2. Development Costs: Unknown