Online Multi-object Visual Tracking using a GM-PHD Filter with Deep Appearance Learning

We propose a new online multi-object visual tracker based on a Gaussian mixture Probability Hypothesis Density (GM-PHD) filter in combination with a similarity Convolutional Neural Network (CNN). The GM-PHD filter estimates the states and cardinality of an unknown and time varying number of targets in the scene handling target birth, death, clutter (false alarms) and missing detections in a unified framework, and has a linear complexity with the number of targets. However, it lacks the identity of targets. We combine spatio-temporal and visual similarities obtained from object bounding boxes and deep CNN appearance features, respectively, to alleviate its shortcoming of labelling targets across frames. We apply this developed method for tracking multiple targets in video sequences acquired under varying environmental conditions and targets density using a tracking-by-detection approach. Finally, we carry out extensive experiments on Multiple Object Tracking 2016 (MOTI6) and 2017 (MOTI7) benchmark datasets and find out that our tracker significantly outperforms several state-of-the-art trackers in terms of tracking accuracy and precision.