1. Develop TRVD (Traffic Rule Violation Detection) system on edge device (dashcam) and achieve 75% precision and 90% recall with rules: running red light, neglect pedestrian, tailgating and harsh cornering. 2. Explore feasibility of tracking algorithm for road scene. 3. Model deployment on Qualcomm chips, including object detection, object tracking and image classification. 4. Develop Android app with AI features, including traffic rule violation detection and driver fatigue detection.

Develop real-time on-device computer vision related algorithm under low computing resource, including: 1. Object Tracking: i. Design feature for low-quality depth information under correlation filter based architecture. ii. Utilize tricks from different research to solve common issues for object tracking, including occlusion, illumination and blurring. 2. Person Re-ID i. Survey and deploy person re-id model for purpose of long-term tracking, achieve 93% top-1 ranking. 3. Design and develop tracking framework, including object tracking, object detection and person re-id.

1. Forecasting the number of enrollment for Taiwan Academy of Banking and Finance, achieve 75% AUC under limited features. 2. Apply discrete analysis like logistic regression to find out the characteristics of enrollment.