Abstract:
Object detection and tracking are few of those fundamental problems of computer vision that are
being tackled from the inception of the idea of autonomous and intelligent systems. Due to the
sheer importance and enormous nature of the problem, a lot of work has been amassed still leaving
room for more. Non-rigid objects such as animal bodies pose a challenge in tracking due to their
deformable nature. The proposed algorithm uses thresholding to identify objects from the scene
and after user input for the object of interest, incorporates a set of rules that are designed to handle
deformations and changes in the object's appearance. These rules are based on the knowledge of
the object's shape and texture, attained through computer vision techniques and they are updated
dynamically to adapt to changes in the object's appearance The use of this approach benefits the
algorithm to be fast and efficient thus easily handles the required online learning and therefore can
be deployed with real-time results on a CPU. The proposed approach is evaluated on a set of
challenging image sequences, and the results show that it outperforms state-of-the-art tracking
methods in terms of accuracy and robustness. Comparison has been done with four other online
trackers and the resulting F1 scores have exceeded all four. The proposed method can be applied
to a wide range of applications such as surveillance, robotics, and human-computer interaction
