Yun-Pu Wu

Wireless connectivity testing solution development.

Computer Vision • Lane departure detection - Developed a pixel-wise conversion algorithm to filter the noise and interpret road semantic segmentation, saving 15% of the time. - Developed the detection algorithm based on the output of the conversion algorithm to cover extreme cases and consider camera installation problems, enhancing overall 20% accuracy. • Facial feature detection - Developed an iris landmark detection model based on Mediapipe, and considered each landmark's confidence value and the computational cost (FLOPs). - Developed a face mask detection feature by extending the network based on the t-SNE analysis of face feature maps from the last few layers. Embedded System • NXP I.MX 8 series - Maintained configurations of Yocto and built the image, and managed the toolchain environment to ensure necessary variables. • Bluetooth low energy (BLE) - Determined the efficient broadcast interval of the MCU to establish a stable connection and reduce power consumption by more than 50%. - Developed Bluetooth features for the Android application to manage multiple devices, and support auto-connect and re-connect.

• Joint Power Control and Rate Adaptation for Vehicular See-Through Systems Proposed an algorithm to address the various uncertainties in VANET by stochastic optimal control techniques.