This project is a vision-based driver drowsiness estimation system from facial expressions. Our system can estimate drowsiness levels accurately, which can prevent drivers from fatigue driving in the early stage. Compared to previous approaches such as Electrogastrogram (EGG) or binary detection CNN models, this system is both user-friendly and accurate.
Bridge inspection is an important issue which usually takes lots of human effort to toil. We used SLAM and reinforcement learning to make the drone plan 3D routes and inspect the bridge automatically. The embedded system of our UAV is the Nvidia Jetson TX2, which can do both SLAM and RL-based planing in real time.
In this project, we built an FCN-style neural network, Taipei Traffic Density Network (TTDN), for analyzing urban traffic. The network generates density maps of vihecle counts from road-side cameras which are usually low-frame-rate, low resolution and low contrast. We use this network to estimates real-time traffic counts of Taipei City, Taiwan.
In modern physical synthesis of VLSI circuits, routability has raised as a primary concern. So far, we have notice that the correlation between global routing (GR) estimation and detail routing (DR) is an essential issue when we want to make a design routable. In this project, We constructed a neural networks to apply supervised learning on congestion estimation. The model can predict the actual congestion and DRC in global routing stage or even embed into placement stage.
In this project, we gathered both students from engineering and designing background to innovate interesting IOT products. We used several tags imbedded on furniture in different houses and design an algorithm to read the communication between the tags. Our goal is trying to create a stress-free relationship between people. Finally, this product was nominated in I Design Award, Taiwan.