Mobile Robotic Surveillance Systems: Detecting and Evaluating Changes in 3D Mapped Environments

This paper presents a study of three possible techniques to detect changes in the environment based on the use of a combination of the Earth Mover’s Distance (EMD) and each one of the following methods: Principal Components Analysis (PCA), Gaussian Mixture Models (GMM) and Plane Extraction (PE). It starts by describing how the EMD can be used with each one of the three methods to detect changes among 3D point clouds that one obtains as robot measurements made with a laser. The advantages of using a laser instead of cameras are related with the sensibility of cameras to weather conditions, while the capability of change detection is crucial for truly autonomous mobile robots, as well as for mobile robots integrated in surveillance systems, for example. The techniques here presented where either implemented in MatLab or adapted from already existing implementations, and a set of simulations were made to evaluate their behaviour relatively to the following aspects: their sensibility to errors in the data, their ability to detect objects of different sizes and their computational complexity. The results obtained show that the combination PCA-EMD is the one with the lowest computational complexity. However, the difference between the PCA-EMD and GMM-EMD techniques is not much significant. More importantly, the results clearly show that the GMM–EMD technique is by far the more stable, in the sense that presents a less sensibility to errors and is able to detect changes with greater reliability. In fact, the techniques PCA-EMD and PE-EMD yield too many fluctuations in the results, making it impossible to distinguish if the technique is detecting a change or the result obtained is because of errors. The GMM-EMD technique has proved to be the best one allowing the detection of changes as was intended in this work. Index Terms Change detection, Earth Mover’s Distance, Principal Components Analysis, Gaussian Mixture Models, Plane Extraction.