Case Study: Computer Vision Engine for Education Technology
“Working with High Dimension was fantastic. After we developed a project plan, they quickly built a working prototype and helped us integrate that into our mobile application.”
Aisha Yusaf, CEO, Pika
Purpose
Our client Pika Vision develops a camera for kids that can recognize objects in the real world. Their innovative solution lets children interact with the application in real time.
The CTO wanted us to bring in our computer vision expertise to lay the groundwork for development and guide their AI strategy. Working on a mobile computer vision application, the key requirements were:
- High accuracy of the machine learning model.
- Running at 30 frames per second (real time) on a mobile device with a high power efficiency.
- Ability to extend the model for future projects.
Approach
The project was done in three stages. In the first stage, the requirements were discussed and a project plan was developed. In the second stage, a dataset was constructed to help assess the performance of different solutions. In the last stage, we developed two solutions, compared their performance and helped the client to integrate the better performing solution into their mobile application.
Results
For the first solution, we built a deep neural network. The network performed well but needed more training data to reach a high enough accuracy. Therefore, we developed a second solution, based on more traditional machine learning techniques such as statistical modelling.
Our second solution proved to be very accurate, and after optimising the code, we were able to run the model on a mobile device in real time using only a fraction of the full CPU power.
We completed the project in the given time frame and agreed to continue working together in the future.