Distributed multi-sensor multi-view fusion based on generalized covariance intersection

Abstract

Distributed multi-target tracking (DMTT) is addressed for sensors having different fields of view (FoVs). The proposed approach is based on the idea of fusing the posterior Probability Hypotheses Densities (PHDs) generated by the sensors on the basis of the local measurements. An efficient and robust distributed fusion algorithm combining the Generalized Covariance Intersection (GCI) rule with a suitable Clustering Algorithm (CA) is proposed. The CA is used to decompose each posterior PHD into well-separated components (clusters). For the commonly detected targets, an efficient parallelized GCI fusion strategy is proposed and analyzed in terms of L 1 error. For the remaining targets, a suitable compensation strategy is adopted so as to counteract the GCI sensitivity to independent detections while reducing the occurrence of false targets. Detailed implementation steps using a Gaussian Mixture (GM) representation of the PHDs are provided. Numerical experiments clearly confirms the effectiveness of the proposed CA-GCI fusion algorithm.

5G mmWave Cooperative Positioning and Mapping Using Multi-Model PHD Filter and Map Fusion

“…On the other hand, AA is more suitable for higher rates of missed detections since it tends to preserve all the detected objects. Taking into account these characteristics, it is clear that GCI fusion is hard to combine with sensors that have limited FoVs [27], [28], because the limited FoVs increase the chances of misdetected objects. For this reason, we choose to use AA.…”

“…Notice that such a region includes the accumulated FoV F k|k (x, m) refer to X k , the PHD D BS k|k−1 (x, m) refers to X k , and finally the PHD D (n) k|k (x, m) refers to X k . Given the decompositions in (28) and (29), the fused PHD is then computed as…”

“…To spread out the complexity over the network, the focus has been on decentralized methods. The most prevalent methods for multi-object density fusion consider generalized covariance intersection (GCI) [27], [28], which amounts to computing the intersection of information among densities, consequently it can not be directly utilized to fuse multi-object densities defined in different FoVs. This difficulty has been overcome in [20], where the PHD of each vehicle is initialized as non-zero throughout the whole area of interest.…”

Recent advances in multisensor multitarget tracking using random finite set

“…One solution is to extend the prior FoVs of all sensors to an identical and larger FoV before fusion (Battistelli et al, 2017;Li SQ et al, 2018), but this may underestimate the existence probabilities of targets in non-public areas. Another widely used method is to split the information according to the FoVs' intersection (Gan et al, 2016;Vasic et al, 2016;Li TC et al, 2019c;Da et al, 2020b;Li GC et al, 2020). Fusion will be executed in each splitting area separately, followed by a reunion operation for all areas.…”

Multi-Source Data Fusion and Target Tracking of Heterogeneous Network Based on Data Mining

“…Li et al [9] took the target and image data as experimental inputs, and observed that not all regions of the image or target are key effective information for cross modal mapping in the generalization problem. Li et al [10] pioneered the application of zero-shot learning (ZSL) and cross domain hash to SBIR, and greatly expanded the research scope of ZS-SBIR. Experimental results show that Li's method only achieved a 1% higher map than traditional methods, but the introduction of the heterogeneous network is a breakthrough.…”