Develop and maintain deep learning models for traffic light recognition system.

Part-time Research Assistant advised by Professor Feng Jiashi Ÿ Designed a speed-efficient algorithm to simulate optical flow features from motion vectors in compressed videosŸ Improved compressed video action recognition pipeline with the designed algorithmŸ Summarized our work in my undergraduate thesis paper

Part-time Research Assistant advised by Professor Yuan JunsongProject: Players and Ball Tracking From Multi-Camera Soccer Videos Ÿ Performed camera calibration for the multicamera systemŸ Re-implemented probability occupancy map for multicamera trackingŸ Redesigned the gridding used in probability occupancy map and improved performance

Project: Vehicle Anomaly Detection for Highway Surveillance Videos (Mar. 2018-May 2018)Ÿ Designed a pipeline to detect anomalous vehicle from highway surveillance videos in real timeŸ Finetuned faster and Mask R-CNN for vehicle detection Ÿ The designed algorithm was able to handle camera jitteringŸ Our pipeline achieved first prize for CVPR Workshop competition, Nvidia AI City Challenge Track2 2018Ÿ Summarized our work as a paper for CVPRWProject: Patient Fall-Down Detection from Hospital Surveillance Scenes (Jan. 2018-Feb. 2018)Ÿ Applied OpenPose to detect human skeletons in the scene Ÿ Designed a classifier to classify skeletons as fall-down or not, which achieved 98% accuracy on test images

Checked foreign exchange balance for client companies, verified clients’ proposals for credits and issued letters of credit for international trading.