Systematic Parameter Optimization and Application of Automated Tracking in Pedestrian-Dominant Situations
Dariush Ettehadieh, Bilal Farooq, Nicolas Saunier · PolyPublie (École Polytechnique de Montréal) · 2015
Though a wealth of data exists for the characterization of pedestrian movement, a majority of said data originates from experimental settings owing to the current state of trackers for real-world scenarios. While these trackers are steadily improving, they remain insufficiently reliable for the accurate, microscopic tracking of individuals, particularly in cases of occlusion or higher density, complex scenes. The authors propose the use of evolution algorithms in the systematic calibration of the parameters of existing trackers in order to further optimize their performance – evaluated by tracking accuracy and precision metrics – in complex cases, with an initial focus on two tracking methods designed for multimodal analysis. Two real test cases were used a) a confined corridor in a public building and b) a subway station entrance during morning rush hour. Current results demonstrate a halving of tracking errors over both default and manually-calibrated parameters, as well as a strong correlation in performance between similar cases. For applications, flow characterization and directional counting are demonstrated.