低複雜度多目標追蹤法及其在汽車雷達之應用

Abstract

在汽車道路環境中，採用交互多模-聯合機率資料相關結合(Interacting Multiple Model-Joint Probabilistic Data Association; IMM-JPDA)技術於汽車雷達追蹤系統可有效的提升系統效能。在本論文中，吾人針對汽車雷達提出低複雜度多目標追蹤演算法。相對於IMM-JPDA演算法，低複雜度多目標追蹤演算法大量的減少關聯事件矩陣(Association Event Matrix)的產生以及簡化邊界關聯事件機率的數學表示式(Marginal Association Event Probability)。為了將軌跡的起始以及終結納入整個追蹤系統中，吾人提出軌跡管理的方法提供紀錄參數給在視野中的每個目標物並且將整個軌跡的生命週期視為一種生滅程序(Birth-Death Process)。為了進一步提升所提出追蹤演算法的效能，吾人提出刪除錯誤觀測(False Measurement)的方法來保護真正的目標物。藉由電腦模擬驗證所提方法可良好運作在汽車雷達之環境。

Adopting the IMM-JPDA technique in automotive radar tracking systems can improve the performance in automotive environments. In this thesis, a low complexity multi-target tracking algorithm is proposed. Compared to the IMM-JPDA algorithm, the proposed algorithm substantially reduces the generation of feasible event matrices and simplifies the equation of marginal association event probability. To take the track initialization and termination into consideration, we propose a track management method which provides recording parameters to the targets in FOV and regards the life of track as a birth-death process. To further improve the performance of the proposed tracking algorithm, we also propose a false measurement deletion method for protecting the true target. Finally, computer simulation results are included to confirm the effectiveness of the proposed method.