Center for Intelligent Robotics
實驗室名稱:人工視覺實驗室
研究專長:深度學習和電腦視覺
實驗室特色:本實驗室之宗旨為精進電腦視覺 (Computer Vision)相關領域之研究與應用,發展方向不僅需契合產業需求,並須挑戰技術瓶頸之突破。
Name of Lab:Artificial Vision Laboratory
Research Interest:Deep Learning and Computer Vision
Characteristics of Lab:The core of this laboratory is to improve the research and application development of computer vision-related fields, which needs to meet the needs of the industry and challenge the breakthrough of technical bottlenecks.
年齡轉換
Age Transformation
隨著深度學習方法的進步,面部年齡轉換算法的發展已成為計算機視覺領域的一個有吸引力的研究主題。給定一張面部圖像作為輸入,面部年齡轉換指的是生成與輸入面部相同身份但年齡較大或較小的面部圖像,同時保持輸入面部的身份特徵。這是一個具有挑戰性的任務,由於面部外觀變化是由於生理老化過程而引起的固有複雜性,與個體的身體狀況、性別、種族和其他因素有關。近年來,基於生成對抗網絡(GAN)的方法、大型面部年齡數據集的可用性以及商業潛力,已經引起越來越多的關注。
Due to the recent progress made by state-of-the-art deep learning approaches, the development of facial age transformation algorithms has become an attractive research topic in the fields of computer vision. Given a face as input, the facial age transformation refers to the generation of facial images for the input face but with older or younger ages in the sense that the identity of the input face can be well preserved. This is a challenging task due to the intrinsic complexity of the facial appearance variation caused by the physical aging process, which can be related to individual physical condition, gender, race and other factors. It has received increasing attention in recent years because of the effectiveness of the Generative Adversarial Network (GAN) based approaches, the availability of large facial age datasets and commercial potentials.
3D 重建
3D Reconstruction
研究團隊開發了一個使用RGBD相機的3D場景重建系統,該系統可以輸入場景的多個視角並將它們合併成真實尺寸的3D場景重建。除了3D場景重建外,該系統還可以重建3D物體。
A 3D scene reconstruction system using RGBD camera is developed which takes in multiple views of the scene and merges them into a real size reconstruction of the 3D scene. In addition to 3D scene reconstruction, the system can also reconstruct 3D objects.
車牌辨識
License Plate Recognition
在Turtlebot上進行車牌識別(LPR)的目的是取代和減少手動操作。該機器人使用Linux中的ROS系統。它使用安裝在Turtlebot上的Kinect來規劃路徑和建立2D地圖。其中最重要的功能是它能夠自動巡航。
LPR on Turtlebot is to replace and reduce manual operation.The robot uses ROS system in Linux.Using Kinect which is set on the Turtlebot to plan path and set 2-D map.The most important thing is that it can automatically cruise.
車牌辨識與追蹤
License Plate Detection and Tracking
系統獲取實時錄影後,將錄影按每個幀切割成照片,並將純運動模型與深度網絡DAN結合,用於車牌檢測和追蹤。
After the system obtains the real-time recording, the recording is cut into photos according to each frame, and the pure motion model is combined with the deep network DAN for license plate detection and tracking.