工業技術研究院

技術簡介

透過RGB相機擷取人體相關之影像，經由濾除背景並取得與人員有關的資訊，相關資訊可透過OpenPose資料庫，取得人體骨骼關節資訊。同時，透過標記人體影像相關資訊，用以訓練行為意圖估測模型。當取得人體骨骼關節資訊，則可帶入模型中進行意圖辨識

Abstract

Use the RGB camera to capture images related to the human body, and obtain information related to personnel by filtering the background. Relevant information can be obtained through the OpenPose database to obtain information on human bones and joints. At the same time, by marking the relevant information of the human body image, it is used to train the behavioral intention estimation model. When the human skeleton and joint information is obtained, it can be brought into the model for intent recognition.

技術規格

‧平均準確率達90.44％ ‧辨識速度約10 FPS

Technical Specification

‧Average accuracy rate of 90.44％ ‧Recognition speed is about 10 FPS

技術特色

‧DeepSort(卡爾曼濾波＋匈牙利法)：濾除背景，取得與人員有關的資訊 ‧OpenPose資料庫：取得人體骨骼關節資訊

應用範圍

人機/多機協作應用

接受技術者具備基礎建議（設備）

‧軟體：Ubuntu 18.04/Windows, CUDA, ROS Melodic Full Version, C＋＋, Python ‧硬體：IPC(最低需求CPU intel i5、DDR4 32GB、SSD 512G、Nvidia RTX2060)

接受技術者具備基礎建議（專業）

熟悉機器人學、座標系轉換、影像處理等知識

聯絡資訊