2021/4/18  
New surveillance camera can tell who in a crowd is not wearing a mask (2021/03/17)
其它新聞 Other News   
  • US, Japan officials underscore importance of peace in Taiwan Strait
  • New defense minister takes questions on China invasion
  • Taiwan can’t afford to donate vaccines to Paraguay: health minister
  • Taiwan forms travel bubble with Palau
  • Postponed Olympic baseball qualifier to start June 16
  • Taiex closes edge down after TSMC shares go ex-dividend
  • Bohemian markets on a cruise ship and Matsu pilgrimages on a plane
  • Three schools send first student to NTU through Star plan
  • Pineapple makes an all-natural cleaning agent, easy to make at home
  • Robot cake maker produces homogenous egg cakes in Taichung
  • 英文 English 
    New surveillance camera can tell who in a crowd is not wearing a mask (2021/03/17)

    A new surveillance camera can tell if pedestrians are social distancing or wearing masks, even when scanning large crowds. The new CCTV system was developed at Chang Gung University's AI Center. Researchers hope it could be used to improve Taiwan's response to public health threats.

    Students come and go in the atrium all day long. But on this surveillance camera, they're marked with red or green boxes. This visual public health surveillance system was developed by the team at Chang Gung University AI Center. The green boxes indicate a mask is worn, but if you don't social distance, a red box replaces it. The system can even tell if you wear a mask incorrectly, exposing your mouth or nose.

    Chen Jen-hui
    Chang Gung University AI Center
    As long as we can collect and analyze data from all the different locations, then we can monitor it. It's much more comprehensive than the spot checks used in the past. That means, as long as you have surveillance camera footage, it can immediately identify.

    It's very different from current identification systems that only monitor one person at a time based on facial profiles. As the center's deputy head Chen Jen-hui陳仁暉told us, identifying objects with footage like this allows you to monitor who is wearing a mask and social distancing. And the key innovation is it can identify multiple people at once with 97% accuracy.

    Yin Shou-kai
    Chang Gung University Computer Science Dept.
    When we first wanted to collect some data, we asked students from the lab to walk around spontaneously, and collected that data.

    Huang Chun-lin
    Chang Gung University student
    Actually we don't have a lot of fancy patterned masks in our dataset, so there could be some errors identifying those. We'll have to adjust it to the most suitable parameters for the dataset.

    They plan to deliberately create more data using multicolored masks to overcome that flaw. In time they hope to use data from public CCTV cameras, which could be used to improve Taiwan's public health system.
    中文 Chinese  
    長庚大學研發視覺防疫系統 口罩沒戴好也能揪出

    後疫情時代,民眾外出得戴上口罩,還是有很多人會忘記,甚至只有經過出入口的時候才意思戴一下。長庚大學師生團隊,研發出一套視覺防疫監控系統,相較現在多數系統只能針對單人辨識。這套系統,包括口罩有沒有正確配戴、有沒有保持社交距離都能掌握,還能多人辨識,準確率高達97%以上。

    入口大廳,學生們進進出出,但從監視器上,怎麼每個人都有紅框、綠框?原來這是長庚大學研發的視覺防疫監控系統,顯示綠色的"MASK"代表有戴口罩,要是沒保持安全距離,則會有大大紅框清楚標示距離,就連口罩沒戴好,露出鼻子或將口罩拉到下巴,也難逃系統法眼。

    [[長庚大學人工智慧研究中心副主任 陳仁暉]]
    "把各個地方監視器畫面,只要能夠蒐集起來通過電腦,那我們就可以去偵測它,這跟過往這種採定點式檢查,又更全面了,也就是說只要有監視器,有影像它就可以做到即時辨識。"

    和現在辨識系統一次偵測一個人,且以頭像為主不同,長庚大學人工智慧研究中心副主任陳仁暉帶領的團隊,運用物件辨識技術,只需要憑著監視影像,不但能掌握有沒有戴好口罩和保持社交距離,重點是可以同時多人辨識,辨識率還高達97%以上。

    [[長庚大學資工系學生 殷壽鎧]]
    "我們一開始就是要做資料蒐集部分,我們是請實驗室的一些同學自然走動,然後我們再蒐集這部分資料。"

    [[長庚大學資工系學生 黃俊霖]]
    ""顏色比較花的(口罩),這個其實並沒有在我們的資料集裡面,可能辨識上稍微有誤差,這個調整我們就是要調整到最適合的數據集的參數。"

    學生還將加強彩色口罩辨識,讓系統更準確,希望未來能與一般路口監視器結合運用,強化全台防疫成效。
    關於民視 常見問題 如何收看民視 意見信箱

    © 2009 FTV All Rights Reserved 版權所有‧盜用必究 Designed by FTV
    地址:244 新北市林口區信義路99號   觀服電話:(02)8601-8601 服務時間:週一至週日09:00-21:00   傳真:(02)8601-3955

    本網站使用 cookie 與其他技術為您提供更好的使用體驗, 請閱讀我們的的隱私權政策,通過使用我們的網站,表示您確認並同意我們的政策。