
Introduction
As North America embraces greener transportation, cycling is expected to become more popular due to its low carbon footprint. However, inadequate road infrastructure and vehicle speeds pose risks to cyclists, leading to accidents and potential injuries, impacting individuals, families, and communities emotionally and physically. Currently, there are dashcams widely available for cars and motorists; however, when it comes to bicycles, there are few. Many out-of-the-box solutions are quite expensive or can be quite difficult to use and maintain.
Objective
Create a mechatronic system that provides reliable rear vehicle detection, convenient mounting and dismounting, and accurate crash identification, allowing drivers and cyclists to share the road responsibly.
Approach
Our project's documentation is located on GitHub. There, we outline our project's problem statement and scope, including goals, requirements, constraints, and hazards. Throughout the project, we conducted verification and validation testing and, in turn, revised and iterated our project's plan and documentation.
We did POCs to multiple stakeholders to ensure usability, resources were being used wisely, and to take in critical and constructive feedback to iterate our product.
Our project is a mechatronic project, specializing in integrating mechanical systems with electronics and software.
- Software - we wanted to build software so robust, systems all around the world could use our software
- Mechanical - we designed and built end-to-end frameworks for the housing and mount using CAD software, 3D printing using Prusa MK3, drilling, sanding, and power tools
- Electrical - we designed and built a unique PCB, wiring and soldering different electrical components like resistors, diodes, capacitors, lights, etc.
Challenges and Lessons Learned
- We adapted our models to improve power and storage efficiency. Initially, we used ML models for our rearview blindspot detection. However, after training our model on thousands of vehicles and presenting a demo of our project at the POC, our model was using way too much power and storage. Therefore, we problem-solved and switched to LiDAR sensors instead. As a result, our software was much more flexible, due to lower compute for the system processes on the Raspberry Pi, and extra access memory on our Pi led to other overall smoother processes.
- We thought outside-the-box to improve energy efficiency. Initially, our product overheated when running all of our modules. However, by multithreading our processes, we were able to use the CPU of the Raspberry Pi to provide multiple concurrent threads to run our processes in parallel. As a result, each function of our safety-critical system could run each function at the correct respective time.
Outcomes and Next Steps
After our final presentation and the Capstone Expo, we came away with a very high mark, exceeding all expectations. Our product was fully functional and we were able to show judges the impact of our work. Furthermore, we were recognized by the James Dyson Award Committee who awarded us as National Runner-Ups for the James Dyson Award. Further details can be found here!
I learned a lot about the importance of version control to maintain documents. By keeping our documents "alive" throughout the year, it allowed for easier revision and collaboration. Furthermore, I was able to apply software engineering concepts like modular design, unit testing, and multithreading I had learned in school to a practical application.
The ideas for future improvements would be as follows:
- Using ML methods for further studies, deeper insights and analytics into the accident.
- The camera data could be used to enhance traffic and safety on the roads, as outlined to us by a GeoTab employee.
- Lastly, we could see cost reduction in various aspects (software, hardware) by having a dedicated chip, containerizing the application as opposed to using a general purpose microprocessor.