我是你的眼
  • 我是你的眼

  • 作品說明

    我們的作品結合了深度學習與邊緣運算裝置,將自動化車牌辨識系統裝在警用機車或汽車車輛上,希望透過系統幫助降低員警因分心造成的行車危險。透過自動化車牌辨識系統即時偵測交通違規行為後,將資訊整合顯示在網頁,並可延伸出其他應用,如開罰違停、贓車歷史定位等。 除了運用在員警巡邏的環境,也可以運用在路邊停車收費的工作中,透過自動化系統來降低所需的大量人力。收費人員僅需拍照後,系統將會自動辨識車牌號碼及偵測路段地點,而該區域拍照完後一次確認欲收費車輛車牌號碼、車種及地點並批次上傳至系統。 我們將攝影機與邊緣運算裝置設置在警用機車與汽車上,攝影機負責接收影像,邊緣運算裝置則負責車牌辨識系統與數據的傳輸,並可透過手機網頁來控制系統的相關設定及查看即時畫面等功能,最後我們會將數據回傳至後台伺服器做數據分析與相關應用。 我們的車牌辨識技術不分晝夜、晴雨、超大角度 (超過 60 度),均能進行有效辨識,在最佳條件辨識率高達 99%,泛用性高,無須多次調整系統參數設定,支援高解析度畫面辨識,即時辨識多個影像來源,單一影像至少可辨識兩車道,並支援嵌入式系統,如:NVIDIA® Jetson Xavier™ NX、NVIDIA® Jetson Xavier™ AGX,且在NVIDIA® Jetson Xavier™ AGX 能夠達到 68 FPS。 Our works install auto license plate recognition system with deep learning on edge computing device and put the device on police motorcycle or police car, hoping to reduce the traffic risk caused by the policeman's distraction. Through real-time detection by auto license plate recognition system, the information will be integrated and displayed on web page, other applications can be extended such as illegal parking, historical positioning of stolen cars, etc. In addition to being used in the police patrol, our works can also be used by roadside parking attendant that reduce the amount of manpower required through our automation system. Roadside parking attendant just need to take pictures of parking vehicle, our system will recognize license plate and get the GPS location. After taking pictures of working region and confirming all vehicle’s license plate, car class and location, then can be uploaded to the system in batches. We set up camera and edge computing device on police motorcycle or police car. Real-time images are received from cameras, edge computing device is responsible for recognition and data transfer. User can see the real-time images and change the setting by mobile web page we developed, the recognized information will send to server for data analysis and other related applications. Our Universal Auto License Plate Recognition System as high as 99%+ accuracy real- time multi-channels in all-weather, unnecessary to adjust parameter setting all the time, robust to all orientations, even more than 60-degree license plate could detection. Also, our system supports high-resolution recognition, real-time recognition of multiple sources, a single image can recognize at least two lanes, feasible on embedded systems like “NVIDIA® Jetson Xavier™ NX” and “NVIDIA® Jetson Xavier™ AGX”. Our system achieves 68 FPS on NVIDIA® Jetson Xavier™ AGX.

  • 隊伍名稱

    All You Need Is Us

  • 指導老師

  • 學校名稱

  • 創作團隊