Selecting and Evaluating Combinatorial Fusion Criteria to Improve Multitarget Tracking

A Review on Intelligent Monitoring and Activity Interpretation

“…We can find many works in literature which propose a series of metrics to evaluate different fusion algorithms, some well-known, such as Euclidean, Mahalanobis or correlation distance, as well as their own metrics used for comparisons in the experiments. We can find works based on genetic algorithms [4], "AND" and "OR" rules [79], or combinatorial analysis [39]. In summary, we can say that the main motivation for performing multisensor fusion is a substantial improvement in the quality of information provided by the sensors, although this improvement is dependent on the fusion technique used according to the input parameters.…”

Combining multiple scoring systems for target tracking using rank–score characteristics

“…More recently in [15][17] [24], the authors have proposed and studied a dynamic and efficient approach to fusion for multitarget tracking in CCTV surveillance (called RAF -Rank and Fuse). Experimental results were obtained to illustrate the use of the RAF approach, explaining the advantage of rank versus score combinations of features for each target.…”

“…We have started a study along these lines ( [15] [43] and protein structure prediction [22]. (3) Frame sequences: In this paper, we applied CFA to each target at each frame, F i , i [1, f].…”

“…Existing approaches to sensory fusion for video tracking have tended to fall into one of three categories: statistical approaches, physical modeling approaches and heuristic approaches. In this paper based on our previous work ( [15]- [17] [24]), we propose a new approach using Combinatorial Fusion Analysis (CFA) (Hsu, Chung and Kristal [14]) which has been applied to other fields such as information retrieval, pattern recognition, virtual screening and drug discovery, and protein structure prediction (see for example, [12] [17]- [18] [22] [26] [28] [43]). This approach is bottom-up and data-driven.…”