Information Fusion-based Road Boundary and Obstacle Detection for Autonomous Vehicles

Abstract

This dissertation described an information fusion-based road boundary and obstacle detection method using three laser radar sensors for autonomous vehicles. Two downward-looking sensors were used to detect road boundaries and obstacles with an estimation of roll and pitch angles of the sensors. The distant obstacles were detected by a forward-looking sensor. In the downward-looking sensors, the algorithm extracts line segments from the raw data of the sensor in polar coordinates. After that, the line segments are classified into road and obstacle line segments. In order to enhance the classification performance, the estimated roll and pitch angles of the sensor relative to the scanning road surface in the previous time step are used. The classified road line segments are applied to track the road boundaries, roll, and pitch angles by using an integrated probabilistic data association filter, which provides both probability of track existence and data association for target tracking in clutter. In addition, information of vehicle navigations and road maps were integrated with the target model. By this information fusion, the detection result of road boundaries and obstacles were improved. In the forward-looking sensor, obstacles are also extracted as line segments. The cluttered measurements corresponding to the scan points on road surfaces due to roll and pitch motions of the vehicle were compensated by the estimated roll and pitch angles of the downward-looking sensors. These roll and pitch compensation were achieved in a synchronized communication by FlexRay-based system. The FlexRay-based network architecture can minimize the worst-case time delay for the sensor fusion. The proposed method was evaluated with the autonomous vehicle A1, which was the winner of 2010 Autonomous Vehicle Competition in Korea organized by the Hyundai Motor Group. The proposed method can accurately detect the road boundaries and roadside as well as road obstacles under various road conditions including paved, unpaved roads, and intersections.|이 논문은 다중 레이저레이더 센서를 이용한 자율주행자동차의 정보융합 기반 도로 경계 및 장애물 검출 알고리즘을 나타내고 있다. 지면을 비스듬히 스캐닝하는 두 개의 레이저레이더 센서는 센서의 롤• 피치각 추정과 함께 도로 경계 및 장애물 검출을 위하여 사용되었으며, 원거리 장애물은 지면과 수평으로 스캐닝하는 레이저레이더 센서를 이용하여 검출하였다. 제안된 알고리즘은 도로 종류에 관계없이 도로 경계 및 장애물을 차량의 국소좌표계 상에서 검출할 수 있다. 이러한 국소좌표계 상의 검출방법은 고성능 GPS와 관성측정장치(IMU)를 이용한 위치인식(Localization)시스템이 불필요하다. 특히, 제안된 알고리즘은 스캐닝한 지면 대비 센서의 롤• 피치각도 추정할 수 있으며, 추정된 롤• 피치각은 지면과 수평으로 스캐닝하는 센서의 차량 롤• 피치로 인해 발생하는 거짓양성 검출 수를 줄일 수 있도록 센서 융합하였다. 이 때, 센서 융합의 효율성을 위하여 FlexRay 기반 시스템 아키텍처를 구성하여 센서 융합을 위한 최악 네트워크 지연시간을 감소시켰다. 또한, 차량 항법과 도로 지도 정보를 융합하여 도로 경계 및 장애물 검출 성능을 개선하였다. 제안된 알고리즘은 한양대학교 자율주행자동차 A1에 적용되어 2010년 현대 자동차 그룹 주관 자율주행자동차 경진대회(AVC)에 출전하여 우승하였다. 제안된 알고리즘은 도로 종류에 관계없이 도로 경계 및 장애물을 정확하게 검출할 수 있다.; This dissertation described an information fusion-based road boundary and obstacle detection method using three laser radar sensors for autonomous vehicles. Two downward-looking sensors were used to detect road boundaries and obstacles with an estimation of roll and pitch angles of the sensors. The distant obstacles were detected by a forward-looking sensor. In the downward-looking sensors, the algorithm extracts line segments from the raw data of the sensor in polar coordinates. After that, the line segments are classified into road and obstacle line segments. In order to enhance the classification performance, the estimated roll and pitch angles of the sensor relative to the scanning road surface in the previous time step are used. The classified road line segments are applied to track the road boundaries, roll, and pitch angles by using an integrated probabilistic data association filter, which provides both probability of track existence and data association for target tracking in clutter. In addition, information of vehicle navigations and road maps were integrated with the target model. By this information fusion, the detection result of road boundaries and obstacles were improved. In the forward-looking sensor, obstacles are also extracted as line segments. The cluttered measurements corresponding to the scan points on road surfaces due to roll and pitch motions of the vehicle were compensated by the estimated roll and pitch angles of the downward-looking sensors. These roll and pitch compensation were achieved in a synchronized communication by FlexRay-based system. The FlexRay-based network architecture can minimize the worst-case time delay for the sensor fusion. The proposed method was evaluated with the autonomous vehicle A1, which was the winner of 2010 Autonomous Vehicle Competition in Korea organized by the Hyundai Motor Group. The proposed method can accurately detect the road boundaries and roadside as well as road obstacles under various road conditions including paved, unpaved roads, and intersections.