A Study on Distributed Multi-Sensor Fusion for Nonlinear Systems Under Non-Overlapping Fields of View.

Area of Science:

• Computer Vision
• Robotics
• Signal Processing

Background:

• Accurate multi-target tracking is crucial for autonomous systems.
• Distributed fusion in asynchronous, nonlinear systems presents challenges due to varying sensor viewpoints.
• Existing methods struggle with nonlinear target dynamics and spatial segmentation.

Purpose of the Study:

• To investigate how different sensor viewpoints impact distributed fusion accuracy in nonlinear multi-target tracking.
• To develop a novel tracking approach that accounts for differential sensor perspectives.
• To improve the fusion of nonlinear moving-target data and spatial segmentation.

Main Methods:

• Proposed a differential-view nonlinear multi-target tracking approach integrating Gaussian mixture, jump Markov nonlinear system, and cardinalized probability hypothesis density (GM-JMNS-CPHD).
• Partitioned observation space by viewpoint boundaries and used TOPSIS-SOS for outlier identification.
• Reconstructed multi-Bernoulli cardinality distribution for target population modeling in subregions.

Main Results:

• The proposed method demonstrated robustness and accuracy in simulations.
• Achieved a lower error rate compared to benchmark algorithms.
• Maintained computational complexity comparable to the standard GM-JMNS-CPHD filter.

Conclusions:

• Differential viewpoints significantly influence distributed fusion accuracy in nonlinear multi-target tracking.
• The proposed approach effectively fuses data and segments targets from varying perspectives.
• This method offers a promising solution for enhanced multi-target tracking in complex visual-field systems.