Taiwan's AI Traffic Enforcement Faces Scrutiny Over Ambulance Exemptions
L'essentiel
- Taiwan's AI-powered traffic enforcement system is under review after drivers were ticketed for yielding to ambulances.
- The Ministry of Transportation stated that human review excludes over 23,000 such cases annually, and AI is being enhanced to improve accuracy and reduce erroneous fines.
Résumé généré par IA
Pourquoi c'est important
Taiwan's AI-powered traffic enforcement system is facing scrutiny due to instances where drivers yielding to ambulances were ticketed. Legislator Wang Yi-chuan raised concerns about the system's limitations in recognizing emergency vehicles, potentially leading to incorrect fines and burdening citizens with appeals.
科技執法普及,屢傳駕駛因禮讓救護車闖紅燈或超越停止線,遭偵測取締,民進黨立委王義川認為,顯示現行科技執法系統於特定情境下,仍可能存在辨識與判讀限制。對此,交通部指出,現行科技執法設備以自動化方式蒐證違規影像,採證影像並須經員警「人工審核」,去年認定屬避讓消防車及救護車等車輛行為而主動排除不予舉發有2萬3749件。
王義川提出書面質詢指出,目前科技執法雖已設有AI辨識與人工複審機制,惟第一線執法人員亦坦言,影像辨識仍可能因拍攝角度、車型差異、環境條件等因素,無法百分之百精準辨識救護車等緊急車輛,後續人工審查亦可能存在疏漏。科技執法既已大量導入AI辨識與自動化判定,主管機關即應持續檢視辨識準確率、人工複審流程與例外情境處理機制,以降低錯誤舉發情形,與民眾承擔申訴與舉證成本。
交通部表示,現行科技執法設備係以自動化方式蒐證違規影像,採證影像並須經員警「人工審核」,若從影像中研判車輛禮讓救護車或其他執行緊急任務車輛,而有交通違規行為,可依法免予舉發,去年全國認定屬避讓消防車及救護車等車輛之行為而主動排除不予舉發之件數計有2萬3749件。
交通部解釋,民眾如因攝影角度或畫面遮蔽等因素,致實際禮讓情形未能即時辨識而遭舉發,得檢具相關佐證資料(如行車紀錄器影像)或敘明違規時間、地點等,向舉發機關提出陳述意見;經查證屬實者,將函請處罰(監理)機關依法撤銷該舉發案件。
交通部指出,今年5月4日函請內政部警政署研議精進科技執法舉發案件審核作業機制,該署並已於5月8日函請各地方政府警察局積極推動相關精進作為,包括推動AI輔助審核機制,將逐步導入AI輔助辨識及舉發前複核功能,協助員警比對影像及主動排除不宜舉發案件,以提升執法精準度。
此外,強化過濾與複核機制,如因攝影角度或畫面遮蔽無法即時辨識者,將由員警進行後端人工複核,確認後依法不予舉發。同時加強後端審核,應主動排除舉發因客觀事實出於不得已之行為,如駕駛人停等紅綠燈時避讓救護車等狀況,以減少民眾需自行申訴之不便。
À surveiller
Perspective IA — des possibilités, pas des certitudes
The AI-assisted review mechanism will be gradually introduced to improve the accuracy of traffic enforcement.
Très probable · En quelques mois
The number of erroneous fines issued for yielding to ambulances will decrease.
Probable · En quelques mois
Questions ouvertes
- What is the current accuracy rate of the AI in identifying emergency vehicles?
- What specific improvements are being implemented in the AI and human review process?
- What is the timeline for the full implementation of the AI-assisted review mechanism?
- How will the system ensure that drivers are not penalized for actions taken out of necessity to yield to emergency vehicles?




