應用於車用雷達與光達的外部及時間延遲校準之高效數據採集方法

儘管車用雷達已經廣泛地使用於自駕車系統，但相比其他感測器，針對雷達與其他感測器之間的校準方法，目前仍然只有少數研究。在本篇論文中，我們提出了一套高效的數據採集方法用以雷達與光達之間外部參數校準及時間延遲校準。對於雷達與3D光達之間進行校準的最大挑戰來自於雷達測量值中偵測物體的仰角缺失。為了解決缺失仰角所帶來的校準困難，整套校準流程被規劃成特徵點的建立以及二階段的優化問題。在進行特徵點的建立中，我們採用建圖-定位的方式從光達的測量值中計算目標物的位置，如此便可解決光達在遠距時因解析度造成的目標錯誤偵測的問題。在二階段的優化問題中，第一階段採用雷達測量值與光達測量值中的幾何資訊作為準則，計算出六個自由度的外部參數以及時間延遲；第二階段則是基於雷達測量值中物體反射強度與其目標物的所在仰角關係提出假設，再去優化第一階段中無法良好估算的參數。為了完成二階段的優化問題，快速運動的目標物觀測以及雷達視野中廣泛仰角的目標物觀測是較為合適的。在這篇研究中，我們提出了一套數據採集的方法去滿足這些數據需求。最後，實驗結果展現了提出方法的可行性以及良好表現

關鍵字

校準

並列摘要

While automotive radars are widely used in most assisted and autonomous drivingsystems, only a few works were proposed to tackle the calibration problems of radars withother perception sensors. In this study, efficient data collection methods for extrinsic andtemporal calibration of radar and 3D LiDAR are proposed. One of the key challengesof calibrating automotive planar radars with 3D LiDAR is the missing elevation anglein 3D space. To address the difficulties caused by the lack of elevation angle of radarmeasurements, the calibration process is characterized as correspondence establishmentand two-stage optimization. In the correspondence establishment, we use a mapping-localization approach to deal with the problem of poor estimation of target center fromLiDAR measurements at long-range targets. In the two-stage optimization, the first stageestimates the six degrees of freedom (DoF) extrinsic parameters and the time delay basedon the geometric constraint between radar and LiDAR measurements; the second stagefurther refine the estimated parameters in the first stage optimization based on the rela-tion between radar’s echo intensity and the elevation angle of target. For accomplishingtwo-stage optimization, the observation of target with rapid motion and the observationof target at a wide range of radar elevation angle are preferable. In this thesis, we suggestefficient data collection methods to satisfy these data requirement. Our experimentalresults show the feasibility and effectiveness of the proposed data collection approach.