Adaptive Spatio-temporal Model Based Multiple Object Tracking in Video Sequences Considering a Moving Camera

Tracking multiple objects in a moving camera is challenging. Due to the irregular movements of the camera, the displacement, scale, and appearance of the objects can be difficult to predict and track. To cope with these problems, we propose an Adaptive Apatio-temporal (AST) model, which explicitly estimate the movement and scale of targets in the view of the moving camera. Moreover, the interactions among objects are also considered to increase the robustness. We introduce our model to the multiple hypothesis tracking and achieve a competitive result on the public benchmark, which includes video of both moving and statistic camera.