Video Analysis and Indoor Sensing

I built a CNN model with YOLO backbone and further designed the centroid algorithm to detect and track pedestrians in real-time
The model detected human figures and their centroid movement in the videos. The model was trained with Triplet loss to re-identify pedestrians and portrait the trajectories.
The algorithm achieved an overall traffic counting accuracy of 77% and was adopted by the univeristy to evaluate the effectiveness of staircase decoration in encouraging students to take more stairs instead of elevators.

Artemis: A Mood Detection and Communication Device

Designed and built the mood detection and communication device, to strengthen the parent-children bounding at long distances. The team consists of engineers from HKUST and designers from the Chinese Art Academy.
Design: I combined the moon phase with the changing mood of people, and used the moon’s metaphor as ‘family union’, hoping to strengthen the communication and bonding within each family.
Technology: I integrated facial expression detection to detect the mood of a person, and use socket transmission to realize the mood exchange.