A new clustering algorithm with adaptive attractor for LIDAR points

Clustering is a semi-supervised or unsupervised algorithm for classifying a set of data according to underlying characteristics or similarity. There are many different algorithms for different applications. Each algorithm has its advantages to some special fields. As to the data obtained from an automotive LUX-LIDAR, the existing algorithms are failed to cluster them accurately or efficiently. It is because that the distances between points are non-uniform and the number of objects is unknown and the object's shape and size are arbitrary. Then, we propose a new clustering algorithm for this kind of LIDAR's sparse data. In the algorithm we introduce a variant described as adaptive attractor. The adaptive attractor is determined by the dataset itself. It determines which class the point belongs to. Compared to the existing clustering algorithm, the advantage of the algorithm lies in the following: i) it is able to classify object with arbitrarily shape and size; ii) the object's number in the ROI is unknown; iii) similarities between data features of different objects are not fixed; iv) its time complexity is low; v) it is simple.