Usage of multiple LiDAR sensors on a mobile system for the detection of persons with implicit shape models

The focus of this paper is the processing of data from multiple LiDAR (light detection and ranging) sensors for the purpose of detecting persons in that data. Many LiDAR sensors (e.g., laser scanners) use a rotating scan head, which makes it difficult to properly time-synchronize multiple of such LiDAR sensors. An improper synchronization between LiDAR sensors causes temporal distortion effects if their data are directly merged. A merging of data is desired, since it could increase the data density and the perceived area. For the usage in person and object detection tasks, we present an alternative which circumvents the problem by performing the merging of multi-sensor data in the voting space of a method that is based on Implicit Shape Models (ISM). Our approach already assumes that there exist some uncertainties in the voting space. Therefore it is robust against additional uncertainties induced by temporal distortions. Unlike many existing approaches for object detection in 3D data, our approach does not rely on a segmentation step in the data preprocessing. We show that our merging of multi-sensor information in voting space has its advantages in comparison to a direct data merging, especially in situations with a lot of distortion effects.