The Cycle Atlanta project aims at creating sensor systems that allow a bike to "see" its environment and collect data as a participatory effort so that we can help the City of Atlanta to make informed decisions about biking infrastructures. Specifically, a sensor box equipped with sonars, lidars, PM sensors, gas sensors, gyroscope, accelerometer, and others was developed to detect environmental factors that can give rise to cyclists' stress level. I participated in this project as a Data Science for Social Good (Atlanta's DSSG) Summer fellow in 2017.
My team mainly worked on (1) reverse-engineering all the prototypes developed previously by the LMC6650 class and consolidating all sensors into two microcontoller units (i.e., Raspberry Pi and Arduino) using the Master-Slave architecture; (2) identifying hardware/software bugs/flaws and refining systems iteratively; (3) implementing fault-tolerent techniques; (4) classifying environmental factors (to detect semantic-level objects) after collecting data using GoPro, voice recorders, and sensors; and (5) visualizing and analyzing the collected data using 3D models and machine learning techniques. My role was to design the system architectures and program on the microcontrollers using Python, C, and shell scripts; to manage the team in developing hardware/software systems, and to analyze data using some signal processing and machine learning techniques (e.g., Fourier/cosine transform, SVM, RandomForest, etc.).
For the initial attempt to predict the environmental factors using sensor-level signals (only using the cosine-transformed, frequency-level signatures), it was possible to predict up to 50% of target objects when formulating this object-detection task as the 7-bin classification problem. Once the prediction power for the environment detection becomes reasonable by engineering features and collecting more data, the sensory data will become eligible to be used for semantic-level empirical studies that identify the relationships between environmental factors and cyclists' stress level.
* I am grateful to have this opportunity to work on this project. This is an on-going project at GeorgiaTech led by our amazing PI, Dr. Christopher Le Dantec.
* Our team members for Atlanta's DSSG 2017 are Javier Argota (CMU), Noel Mannariat (Mahindra Ecole Centrale), Erica Pantoja (Kennesaw State), and myself.