On the Detection of Moving Objects in Laser Scan Data: the Highest Point of Interest (HPOI) Method

Abstract

There are many sensors and measuring methods for detecting moving objects, each with its advantages and disadvantages. In active tracking methods (based on e.g. GNSS technology), the user is informed and actively participates, for instance by installing a smartphone app. These methods typically have the problem that only a limited part of the moving objects is tracked. In passive tracking methods (e.g. video recognition), the moving person is not informed of being subject to the data acquisition. These methods are typically privacy-invasive. Many techniques also require complex calculations to transform the raw data into accurate and meaningful trajectories of moving objects. However, such trajectories usually require only one point of the moving object at any given time. If the moving object is a person walking or cycling, then such a point of interest is the highest point of the person's head (i.e. “highest point of interest” or HPOI). Detecting this point typically demands computationally intensive mining of the trajectory data, for example using deep learning approaches in video recognition. We present the use of static LiDAR technology, a well-established, precise and anonymous 3D data acquisition method, for this use case. By continuously (i.e. at a high temporal rate) laser scanning an environment in which pedestrians or cyclists move, multiple epochs of point clouds are obtained. A robust vertical threshold filtering allows reducing aforementioned high-dimensional, bulky point cloud data to easily visualisable and interpretable trajectories of HPOIs.