Recognition And Location Prediction Of Interactive Groups In A Large Shopping Mall

In public places,grouping of people is a very common phenomenon,such as shopping at shopping malls or walking in the park with friends or relatives.Interactive groups refer to the groups whose members have interactions such as waving hands,shaking hands,embracing,or walking hand in hand,which are not uncommon occurrences in our daily life.Existing group recognition approaches are based on the similarity of the individuals' locations or signal features.However,in the real life,the members of the group are not always close to each other and keep the walking speed and direction similar.The interactions among people are probably regarded as dissimilar and affect the recognition accuracy.In this paper,we propose an approach called Interactive Group Recognizing(IGR)for recognizing groups with interactions among their members.The acceleration data of individuals are collected,and their actions are inferred based on the data.The disparity between two individuals is computed using the sliding window technique.We further recognize the groups using a majority-voting based method in order to solve the problem in which not all group members perform the same interaction at the same time.Compared with the existing approach,the average group recognition accuracy and Fl-score is improved by 6.9%and 13.6%.Our experimental results also show that the average accuracy and F1-score of IGR can reach 97.2%and 94.7%when interactions among people account for no less than 8%of the total actions.In the interactive group location prediction,the sequential tree storage structure is proposed in consideration of the incremental update of the dataset.This structure can obtain the frequent area sequences and the corresponding association rules by scanning the database only once.When the data set is dynamically increased only need to add the newly added data set to the existing sequence tree to do incremental updates.Due to the nature of the sequence tree structure,the process of predicting the group position is the traversal process of the tree structure.Compared with other data storage structures,the tree structure has significant advantages in the traversal efficiency.At the same time,combining the association rules stored in the corresponding tree nodes,we can obtain single-step and multi-step position prediction results in the traversal process.Finally,in order to improve the accuracy of group location prediction,a method based on group appearance time and group number is proposed.We implement experiments on real dataset.Compared with the control algorithm,the proposed method has a higher accuracy of position prediction,and the accuracy of single-step and multi-step position prediction can reach 91.2%and 33.8%respectively.