本文为瑞典查尔姆斯理工大学（作者：JOHAN VILLYSSON）的硕士论文，共53页。

汽车主动安全领域是当今快速发展的领域之一。近年来，更安全的汽车设计已成为汽车制造商的首要任务之一。使汽车更安全的驾驶员支持功能依赖于汽车周围精确的车载传感器数据。为了验证这种传感精度，可以使用激光雷达获得车辆周围360度的视图，然后将其用作地面真实度的观测依据。尽管激光雷达具有很高的精度和分辨率，还有典型的宽视场，但从激光雷达数据（即所谓的点云）中提取物体需要先进的后处理算法，而且往往具有挑战性。本文提出并实现了一种利用激光雷达点云进行动、静态目标检测与跟踪的离线跟踪算法，该跟踪器以贝叶斯跟踪理论为基础，利用了前向和后向跟踪都是离线处理的特点。跟踪器的结果显示了与文献中现有跟踪器相当的精度。该跟踪器能够有效地跟踪任何类型的物体，无论它们是静止的还是动态的，并给出位置、尺寸、速度和加速度的精确读数。

The field of active safety in theautomotive industry is a fast-growing area of development today. Safer carshave become one of the top-priorities among car-makers in the recent years. Thedriver support functions that make cars safer rely on accurate onboard sensorreadings of the surroundings of the car. For verification of this accuracy, aLIDAR can be used to get a 360◦ view of the surroundings of the vehicle thatcan then be used as ground-truth. Despite high accuracy and resolution combinedwith wide field of view typical for LIDARs, extraction of objects from theLIDAR data, the so called point clouds, requires advanced post-processingalgorithms and is often challenging. In this thesis, an off-line trackingalgorithm that detects and tracks dynamic and stationary objects using LIDARpoint clouds is developed and implemented. The tracker is based on Bayesiantracking theory and utilizes the fact that it is run off-line by tracking bothforward and backward in time. The results of the tracker show comparableaccuracy to current trackers in the literature. The tracker effectively tracksany sort of objects whether they are stationary or dynamic and gives accuratereadings on positions, dimensions, velocities and accelerations.

1. 引言

2. 跟踪理论

3. 数据与方法

4. 结果

5. 讨论

6. 结论

附录 分割

更多精彩文章请关注公众号：