Improved Agglomerative Information Bottleneck Based Trajectory Clustering And Visualization

Trajectory clustering can dig out some implicit information from the motion trajectory data.When clustering trajectories,most of the clustering algorithms need to specify the final number of trajectory clusters,and the appropriate similarity measure functions need to be selected to measure the similarity between trajectory data.The Agglomerative information bottleneck algorithm introduces trajectory feature space set when clustering trajectory data,and then minimizes the mutual information between the original trajectory data set and the trajectory feature space set,and at the same time,let the clustered trajectory set and the mutual information of trajectory feature space sets is as large as possible.The algorithm can get a good clustering effect under the two constraints of the mutual information.The Agglomerative information bottleneck algorithm is an iterative calculation method.At each iteration,two trajectory clusters that can minimize the objective function are selected for merging.This calculation mode may cause the local optimal solution.We balance the best-case solution by adding a global quality term to the objective function of the cohesive information bottleneck algorithm.Through the experimental results on the synthetic data and the simulated data set,we can find that our improvement on the condensed information bottleneck algorithm is very effective.In order to help the user to understand the concrete computing process of clustering of motion trajectory by the condensed information bottleneck algorithm,we design a trajectory cluster visualization analysis tool.The visualization tool mainly consists of two parts,visualization of trajectory modeling and visualization of trajectory clustering.Through the interactive way to allow users to control the trajectory modeling and clustering process,user-friendly view of the details of the algorithm to deepen the understanding of the algorithm.