Personal Weekly Update | Zoe | 4/26
Accomplishments
This week I wrapped up the final data analysis and experimentation for the project. I finished the main set of slides for our final presentation, creating charts, tables, and summaries based on the completed testing and system behavior.
I worked through all of the calibration and ride data we recorded, analyzing how well the calibration masks and pose matching performed. Based on the results, I finalized the new matching approach that uses both sensor activation masking and intensity scoring. This added step improves pose classification, especially when rider movement is more dynamic during actual pedaling.
I also helped Carolyn finish testing the app’s BLE connection to make sure calibration, riding, and pose feedback all worked as expected when paired with the final system.
Below are two figures showing results from the final tests:
Distribution of Alert Scores by Pose
Calibration Summary: MAE and RMSE Across Poses
Progress Status
The main sensing system, BLE communication, calibration logic, and app integration are all complete and tested. However, the Raspberry Pi imaging setup is still being worked on. The Pi itself and the camera connection are functional, but camera tuning and integration into the full system are still in progress and will continue after the main demo.
Next Steps
– Deliver final presentation and demo
– Submit final project report, code, and system documentation
– Continue working on Raspberry Pi camera setup and basic image tracking integration
Unit Tests and Overall System Tests
Unit Tests performed:
– sensor tuning tests, verifying stable force sensor output with tuned resistors
– BLE communication tests, checking characteristic matching and packet stability
– calibration mask tests, verifying live pose matching during static and dynamic riding
– intensity matching tests, confirming improved pose classification with variable force input
Overall System Tests performed:
– full calibration and riding trials with 13 sensors mounted
– riding tests with bike both mounted on a stand and free riding off the stand
– cross-device BLE testing with multiple phones
– real-time pose detection tests while pedaling and shifting weight
Findings and design changes:
– basic on-off sensor masking was not reliable enough for fast dynamic riding, especially when weight was shared between sensors
– added intensity matching during calibration and riding, which improved pose matching confidence and reduced false positives
– tuning of resistor values on seat and handlebar sensors helped distinguish between seated and leaning poses
– still pose matching during active riding showed slightly higher error (as seen in the MAE and RMSE summary) but remained within acceptable performance
The system consistently detects calibration poses and active riding postures with reasonable accuracy, supports real-time feedback over BLE, and integrates smoothly with the mobile app. Pi camera imaging will continue to be developed to add an additional tracking component.