스마트폰에서 Vision/GPS/INS 센서 퓨전을 이용한 차량 측위 시스템에 관한 연구

Abstract

본 논문에서는 스마트폰 내장 GPS, INS(Accelerometer, Gyroscope), Camera 센서에서 얻은 정보를 확장 칼만 필터를 이용해 융합하여 도심지와 터널과 같은 GPS 음영지역에서 차량의 위치를 추정하는 시스템을 설계하고 그 성능을 분석한다. 음영지역은 GPS NMEA 메시지(GPGSV, GPGSA format)에서 파싱한 SNR(Signal-to-Noise Ratio), DOP(Dilution of Precision) 정보를 활용하여 판단하며 비음영지역 또는 음영지역에서의 측위는 GPS의 경도와 위도, INS에서의 가속도와 자세각, Google ARCore에서 추출한 변위 정보를 확장 칼만 필터에 입력하여 최종 경도와 위도를 추정한다.

시스템의 설계 및 평가는 음영지역이 잦은 도심지와 터널 시나리오로 실차환경에서 진행이 된다. 실차 테스트를 통해 스마트폰 각 센서의 raw data를 취득하고 취득한 raw data를 바탕으로 음영지역 판단 알고리즘 및 확장 칼만 필터의 로직을 설계한다. 음영지역 판단 알고리즘은 음영지역에서의 SNR 값, DOP 값을 분석하여 threshold 값을 설정함으로써 설계된다. 확장 칼만 필터는 가속도(X, Y, Z), 속도(X, Y), 위치(X, Y), Yaw 각, Pitch 각, 헤딩 값의 상태변수를 가지는 10차로 이루어지며 가속도를 업데이트하는 식을 설계할 때 주행가속도 뿐만 아니라 중력 가속도와 바퀴 마찰에 의한 힘을 모두 고려하는 모델을 사용한다. 확장 칼만 필터로부터 최종적으로 나온 경도와 위도는 GIS 소프트웨어인 Global Mapper를 이용하여 위성 지도상에 매핑하고 실제 주행한 경로와 비교하여 오차를 시각적으로 확인하고 측위 결과가 주행 차선을 구분할 수 있는지 확인한다. 시스템의 정량적 평가를 위해서는 기존 GPS 단독 측위와 제안한 시스템의 평균 오차와 90% CEP 오차를 도출하고 비교함으로써 제안하는 측위 시스템의 성능을 확인하고 평가한다.| In this thesis, we design a system that estimates the location of vehicles in GPS shadow areas such as urban areas and tunnels by fusing information from smartphone built-in GPS, INS (Accelerometer, Gyroscope), and Camera sensors using an Extended Kalman Filter. The shadow area is determined by using the Signal-to-Noise Ratio (SNR) and the Dilution of Precision (DOP) information parsed from the GPS NMEA messages (GPGSV, GPGSA format). The final longitude and latitude are estimated by inputting GPS longitude and latitude, acceleration and attitude angles from INS, and displacement information extracted from Google ARCore into the Extended Kalman Filter.

The design and evaluation of the system is carried out in the actual vehicle environment in the urban and tunnel scenarios where shadow areas are frequent. Through the real vehicle test, we acquire the raw data of each sensor of the smartphone and design the logic of the shadow area judgment algorithm and the Extended Kalman Filter based on the obtained raw data. The shadow area judgment algorithm is designed by analyzing SNR value and DOP value in the shadow area and setting the threshold value. Extended Kalman Filter consists of tenth order with the state variables of acceleration (X, Y, Z), velocity (X, Y), position (X, Y), Yaw angle, Pitch angle and heading value. The design uses a model that considers both gravity acceleration and wheel friction as well as driving acceleration. The final longitude and latitude from the Extended Kalman Filter are mapped onto the satellite map using GIS software Global Mapper and compared to the actual route to visually identify the error and verify that the positioning results can differentiate the driving lane. For the quantitative evaluation of the system, we verify and evaluate the performance of the proposed positioning system by deriving and comparing the average error and 90% CEP error of the existing GPS single positioning and the proposed system.; In this thesis, we design a system that estimates the location of vehicles in GPS shadow areas such as urban areas and tunnels by fusing information from smartphone built-in GPS, INS (Accelerometer, Gyroscope), and Camera sensors using an Extended Kalman Filter. The shadow area is determined by using the Signal-to-Noise Ratio (SNR) and the Dilution of Precision (DOP) information parsed from the GPS NMEA messages (GPGSV, GPGSA format). The final longitude and latitude are estimated by inputting GPS longitude and latitude, acceleration and attitude angles from INS, and displacement information extracted from Google ARCore into the Extended Kalman Filter.

The design and evaluation of the system is carried out in the actual vehicle environment in the urban and tunnel scenarios where shadow areas are frequent. Through the real vehicle test, we acquire the raw data of each sensor of the smartphone and design the logic of the shadow area judgment algorithm and the Extended Kalman Filter based on the obtained raw data. The shadow area judgment algorithm is designed by analyzing SNR value and DOP value in the shadow area and setting the threshold value. Extended Kalman Filter consists of tenth order with the state variables of acceleration (X, Y, Z), velocity (X, Y), position (X, Y), Yaw angle, Pitch angle and heading value. The design uses a model that considers both gravity acceleration and wheel friction as well as driving acceleration. The final longitude and latitude from the Extended Kalman Filter are mapped onto the satellite map using GIS software Global Mapper and compared to the actual route to visually identify the error and verify that the positioning results can differentiate the driving lane. For the quantitative evaluation of the system, we verify and evaluate the performance of the proposed positioning system by deriving and comparing the average error and 90% CEP error of the existing GPS single positioning and the proposed system.