SCC/AEB 레이더의 위험 표적 식별 알고리즘 개발

Abstract

레이더(Radar)는 표적과의 거리 및 위치를 측정하기 위해 개발되었다. 그러나 기본적으로 False Alarm과 Mis-Detection의 문제가 발생한다. False Alarm이란 실제로는 표적이 없는데 레이더에서는 표적이 있는 것으로 판단하는 것이고, Mis-Detection은 실제로는 표적이 있는데 레이더에서는 표적이 없는 것으로 판단하는 것이다. 이는 차량용 레이더인 FMCW(Frequency Modulation Continuous Wave)에서도 문제가 된다. 2014년 유럽 자동차 안전성 평가기관인 Euro NCAP(New Car Assessment Program)에서 AEB(Autonomous Emergency Braking)항목을 추가한 이후, SCC(Smart Cruise Control)/AEB가 적용된 차량이 늘어나고 있다. 또한 적용된 차량은 AEB 테스트 방법을 만족해야 한다. 테스트 방법은 CCRs/CCRm/CCRb(Car to Car Rear stationary/moving/braking) 3가지가 있다. CCRs는 전방의 Target Vehicle이 정지해있는 상황에서 AEB 시스템을 만족하는 것이고, CCRm은 전방의 Target Vehicle이 등속 운전하고 있는 상황에서 AEB 시스템을 만족하는 것이다. 마지막으로, CCRb는 전방의 Target Vehicle이 등감속도 운전을 하고 있는 상황에서 AEB 시스템을 만족하는 것이다. 차량용 레이더가 장착된 차량이 전방의 정지 차량을 만난 경우에는 차량으로 인식해서 멈추어야 하고, 전방의 바닥 구조물을 만난 경우에는 단순한 바닥 구조물로 인식하고 그대로 주행해야 한다. 하지만 현재는 AEB 시스템의 인식과정에서 문제점이 나타난다. 그래서 연구의 범위를 AEB 시스템에서 CCRs로 한정한 후, 이에 대하여 새로운 알고리즘을 제시한다. 새로운 알고리즘은 머신러닝 기법 중 PCA(Principal Component Analysis)와 SVM(Support Vector Machine)을 이용한다. 마지막으로 새로운 알고리즘의 성능을 검증하기 위하여 실험 데이터를 그대로 SVM과정을 거친 인식 정확도, 실험 데이터에 처리를 한 후 SVM과정을 거친 인식 정확도, 새로운 알고리즘인 실험 데이터에 처리를 한 후 PCA와 SVM과정을 거친 인식 정확도를 비교한다.| Radar is developed to measure the range and the position of the target. However, during the measuring process there might be problems such as ‘False Alarm’, ‘Mis-Detection’. ‘False Alarm’ means misjudgement of detecting a target when there are actually nothing in the observing range. While ‘Mis-Detection’, on the contrary, is the case that the radar could not detect the targets exist in the observing range. These problems also happen in the case of FMCW(Frequency Modulation Continuous Wave) radars which are used in many cars. In 2014, Euro NCAP(New Car Assessment Program) adopted AEB(Autonomous Emergency Braking). After that the number of cars applying SCC/AEB keeps growing. These cars need to meet the requirements of AEB test method. These are 3 kinds of test methods, CCRs/CCRm/CCRb(Car to Car Rear stationary/moving/braking). These methods means AEB has to meet different conditions that when the front target vehicle is stationary(CCRs), or moving at same speed of user’s car(CCRm), or doing constant deceleration(CCRb). Therefore, cars applying with AEB (FMCW radars) should stop going if there is a resting car in the front, and should keep moving when detected obstacles on the ground. However, conventional AEB has some problems in this detecting process. In this paper, we suggest new algorithm to solve the problems of CCRs of AEB. The new algorithm uses PCA(Principal Component Analysis) and SVM(Support Vector Machine). In the last part, to verify the algorithm, we also made a comparison of the accuracies of these three cases : 1. Applying experimental data directly to SVM. 2. Applying experimental data to SVM after specific data processing. 3. The algorithm in this paper which applying experimental data to PCA and SVM after specific data processing. And by comparing the three cases we can conclude that using the algorithm proposed in this paper has a quite high accuracy.; Radar is developed to measure the range and the position of the target. However, during the measuring process there might be problems such as ‘False Alarm’, ‘Mis-Detection’. ‘False Alarm’ means misjudgement of detecting a target when there are actually nothing in the observing range. While ‘Mis-Detection’, on the contrary, is the case that the radar could not detect the targets exist in the observing range. These problems also happen in the case of FMCW(Frequency Modulation Continuous Wave) radars which are used in many cars. In 2014, Euro NCAP(New Car Assessment Program) adopted AEB(Autonomous Emergency Braking). After that the number of cars applying SCC/AEB keeps growing. These cars need to meet the requirements of AEB test method. These are 3 kinds of test methods, CCRs/CCRm/CCRb(Car to Car Rear stationary/moving/braking). These methods means AEB has to meet different conditions that when the front target vehicle is stationary(CCRs), or moving at same speed of user’s car(CCRm), or doing constant deceleration(CCRb). Therefore, cars applying with AEB (FMCW radars) should stop going if there is a resting car in the front, and should keep moving when detected obstacles on the ground. However, conventional AEB has some problems in this detecting process. In this paper, we suggest new algorithm to solve the problems of CCRs of AEB. The new algorithm uses PCA(Principal Component Analysis) and SVM(Support Vector Machine). In the last part, to verify the algorithm, we also made a comparison of the accuracies of these three cases : 1. Applying experimental data directly to SVM. 2. Applying experimental data to SVM after specific data processing. 3. The algorithm in this paper which applying experimental data to PCA and SVM after specific data processing. And by comparing the three cases we can conclude that using the algorithm proposed in this paper has a quite high accuracy.