Multi-target tracking by learning local-to-global trajectory models

Expert video-surveillance system for real-time detection of suspicious behaviors in shopping malls

“…We achieve an 86.7% in MT, while the best state-of-the-art tracking methods recently published obtain a slightly lower precision of 86.4% (Chau et al, 2014a;Badie & Bremont, 2014). Besides, our algorithm also has a remarkable processing speed near to 50 fps, while the fastest method in the recent state of the art obtained a speed of approximately 31 fps (Zhang et al, 2015). As deduced from the theoretical explanations given in previous sections, the successful performance of our method for people tracking is due to the usage of an association based on Kalman filtering and a LSAP optimization combined with an occlusion management that uses visual appearance based on features such as GCH, LBP and HOG.…”

“…There are several tracking solutions for videosurveillance, and their applicability depends on the scenario: traffic situations (Alvarez et al, 2014), airports (Besada et al, 2005), maritime surveillance (Szpak & Tapamo, 2011), sports events (Kayumbi et al, 2008), places in poor lighting conditions (Wong et al, 2009;Gade & Moeslund, 2014), crowded environments (Zhao & Nevatia, 2004;Wu & Nevatia, 2006;Li et al, 2008Li et al, , 2009Xing et al, 2009;Kuo et al, 2010;Kuo & Nevatia, 2011;Ali & Dailey, 2012;Chau et al, 2014a,b;Badie & Bremont, 2014;Walia & Kapoor, 2014;Guan & Huang, 2015;Zhang et al, 2015), and so forth. However, there is not any realistic approach in related works for conveniently solving the problem of human tracking in a real-time video-surveillance system for shopping malls.…”

“…Currently, human tracking in crowded environments is a research topic in a constant evolution and, apart from the papers previously analyzed in Table 3 (2015); Zhang et al (2015). In these recent works, tracking evaluation is not focused on occlusion situations, as seen in the results of Table 4 using these metrics based on trajectories in order to compare the performance of our approach against the newest tracking algorithms appeared in the state of the art.…”

Ordered trajectories for human action recognition with large number of classes

“…The main challenges include missed detections, false alarms, inaccurate detections, occlusions, and similar appearance among multiple objects. Three popular strategies [27] in use to counter these challenges are: local data association, global data association, and hierarchical data association methods.…”

“…A particle filter was used to generate a set of reliable trajectories in the local stage and a modified Hungarian algorithm was used to optimise the data association in the global stage. Similarly, [27] proposed a unified framework for automatically relearning from local to global information. The local-to-global trajectory models were used to link detections from consecutive frames into trajectories and also link separated trajectories that belong to the same targets into long trajectories.…”