知識庫

Noise Filtering in Autonomous Emergency Braking Systems with Sensor Fusions (具感知融合之自動緊急煞車系統的雜訊過濾器設計)
  • 發布年度:2015
  • 主要類別:電動車與車輛電子
  • 次要類別:論文
  •  

    此篇論文主要著重於具備攝影機與毫米波雷達感知融合之自動緊急煞車系統(autonomous emergency braking system, AEB system)的雜訊過濾設計。文中考慮三種雜訊,分別為兩倍諧波雜訊、地面金屬訊號與鏡面效應。第一種雜訊只要是由於雷達波在前車與本車間來回反射次數多一倍所造成。因此,偵測到的雷達障礙物中會有一個的偵測距離約為另一個的兩倍。第二種雜訊主要是由於雷達偵測到地面物體後所造成。於此狀況下,當攝影機打算抓取前方車輛計算其感測距離時,無法正確抓取到地面物體來計算;反之會改針對更前面的類車物體來做計算。如此一來會造成地面物體的雷達偵測距離會小於其影像偵測距離,可依靠此特點來辨識此種雜訊並過濾之。第三種雜訊乃是由於雷達波偵測到目標車後,回傳過程中又跟周遭環境產生反射導致。此種雜訊具備一種特徵:針對此時偵測到的障礙物位置和障礙位置估測器估算的障礙物位置之間的誤差量做標準差運算後,會發現此值會遠大於非雜訊障礙物的標準差。固可將具備此特色的障礙物視為雜訊障礙物,並過濾之。最後的驗證實驗可顯示出藉由運行本文所提出的過濾器,可有效簡化雷達所偵測到的環境資料,此得雷達可於車速10 km/h到50 km/h之間成功抓取位於前方5到50公尺的障礙物,不因雜訊影響而失敗。
     
    This paper discusses noise filtering in an autonomous emergency braking (AEB) system with a sensor fusion between a millimeter wave (MMW) radar and a camera. Three kinds of noise, namely twice harmonic noise, ground noise, and specular reflection noise, are then filtered. The former is caused by the reflection of a radar wave between a target object and the MMW radar; therefore, one of the sensing distances would be twice as longer as one of others. An object featuring this characteristic is treated as the noise and filtered. Next, detecting a ground metal as the target object generates the second noise with a focus of car-like objects. That is, an object—with the sensing distance from the MMW radar being smaller than that from the camera by a threshold value—is taken as the ground metal noise and ignored. Moreover, the third noise happens when there is a radar wave reflection between an object and its surroundings. While one focuses on standard derivation (SD) of a difference between the real and estimated object position from an object tracker, the specular reflection noise naturally reveals high SD. An object revealing this characteristic is hence removed from sensing data. Verification results indicate that the surroundings filtered by the proposed mechanisms are simplified such that the AEB system catches the right target object with its sensing distance being within [5, 50] (m) and our car speed lying within [10, 50] (km/h).