Jeffrey Johnson Email & Phone Number

Cambridge, Ma, Us

Cambridge, Ma, Us

• Implemented two ports of AOSP (Android Open Source Project) software to NXP ARM Targets. Both targets were versions of TransAct’s BOHA! Terminal series. First port was Android Oreo version to 32 bit ARM. Second port was Android 10 to 64 bit. Test Android applications created. Fixed network related bugs in older terminal version. • Created Android applications which controlled and configured embedded Linux Work Station device. Android work included UI/UX, robust Bluetooth and WiFi network management. Uniform exception based error handling. Changes were made to the Linux Work Station device to facilitate configuration. HTTP. REST server/client model.

• Multiple company clients and Signal Processing/Neural Network research. • Figur8: Created wxPython based visualization tool that synchronized replay of sport video activities with collected accelerometer data. Any subset of graphs of linear inertial and angular inertial axes or their 3D vector lengths could be replayed or statically displayed in a window. Stretch band data was similarly displayed. Band was part of device along with accelerometer. Alternatively, a window could replay a video synchronized to the graphs. In addition to replay at chosen speed, graphs could be magnified or statically navigated to display regions of graphs or video stills of interest. Support for captioning videos or labeling windows was supported. • TrueMotion: Created Android applications for data logging and tracking power usage by TrueMotion background driver safety monitoring application. Ran on Android. • Research: Development of alternative Neural Network convolution algorithms. Algorithms were lower compute cost than standard Neural Networks. Suitable for TinyML. Gave paid Neural Network and accelerometer presentations.

• Developed signal processing algorithms for the detection of Activities (biking, walking, etc.) and Gestures (wrist double taps, waves, etc.). Targets were desktop (Linux and macOS), Android and embedded ARM wearable (Google Jacquard Tag device). The tag is located in the wrist of the jacket. The tag measures and stores limited amounts of calculated algorithmic values, accelerometer data, and input from Jacquard smart fabric of the jacket arm. Activity detection facilitates monitoring exercise activity. Both Activity and Gesture detection facilitate control of an Android phone bypassing phone touch control. • Algorithms were delivered as source in two forms: (1) For the desktop Linux and macOS emulation environments; and (2) For embedded RTOS ARM wearable tag device. C code for the ARM wearable ran with no source changes within the desktop environment. Excluding the embedded C code, desktop code was a mixture of Python and C++. • Driven by data logged from the tag, all embedded algorithms could be run as an emulation in the visualization environment provided by the desktop software. Visualization features included data display, event markings of where activities start and stop and where gestures are detected. Time dependent internal calculations associated with the algorithms could also be displayed. Algorithms are parameterized. Display and automated data analysis methods assisted with the tuning of event detection for Google’s Jacquard Biker Jacket Product. • Algorithm detected event times in the data log were compared against synchronized GoPro video data and human markings of the same events. Visualization software included the simultaneous display of both algorithmic and human detection of events and data. Display included summaries of algorithmic accuracy compared to human classification – often on a second viewing, the algorithm seemed more correct.