During parking, the vision-based parking slot detection system has error or missed detection due to distortion, illumination variation, occlusion and limited field of view (FOV). Thus, the parking slot position sent to path planning system is inaccurate and discontinuous. Besides, the intelligent parking system (IPS) primarily relies on dead reckoning to send vehicle position to the motion control system, whereas the error will accumulate with the rise in driving distance. All the mentioned factors will cause parking deviation, incline or even line-pressing. In this paper, the idea of visual simultaneous localization and mapping (SLAM) is adopted innovatively to achieve the parking slot tracking. Moreover, the extended Kalman filter (EKF) is used to achieve the fusion of vision and vehicle chassis information, to solve the two problems of discontinuous and inaccurate parking slot detection, as well as cumulative error in dead reckoning. Finally, the effectiveness of the proposed algorithm is verified using the vehicle tests.

中文翻译：

---

基于视觉与车辆底盘信息融合的停车位跟踪算法研究

在停车期间，基于视觉的停车位检测系统由于失真，照明变化，遮挡和有限的视场（FOV）而出现错误或漏检。因此，发送到路径规划系统的停车位位置是不准确且不连续的。此外，智能停车系统（IPS）主要依靠航位推算将车辆位置发送到运动控制系统，而误差会随着行驶距离的增加而累积。所有上述因素都会导致停车偏差，倾斜甚至压线。在本文中，视觉同步定位和映射（SLAM）的思想被创新地采用以实现停车位跟踪。此外，扩展的卡尔曼滤波器（EKF）用于实现视觉和车辆底盘信息的融合，解决了停车位检测不连续和不正确的两个问题，以及航位推算中的累积误差。最后，通过车辆测试验证了所提算法的有效性。