An adaptive fusion method used in forward looking sonar multi-feature tracking

In order to improve the accuracy of underwater multi-object tracking based on the forward looking sonar, on the basis of particle filter tracking, the multi-feature adaptive clue fusion method was used to switch fusion methods by adjusting features online to calculate the particle weight. Particles were initialized. Then multiple features of the template corresponding to every particle were extracted, including the basic object shape and intensity features, digital features of moment invariants and digital features of the gray level co-occurrence matrix. The final particle weight was obtained by fusing every feature weight using the adaptive fusion method. Multiplicative fusion was adopted when the features worked well; otherwise weighted sum fusion based on fuzzy logic was adopted. Sequence images through a tank experiment were used to verify the effects of the adaptive fusion method, in contrast to the traditional fusion methods. The images were obtained by using the forward looking sonar, describing two cross motions. The tracking ability using the adaptive fusion method was found to be better. This method has significant effectiveness for automatically tracking autonomous underwater vehicles.