Multiple Object Tracking Based on Faster-RCNN Detector and KCF Tracker

Tracking and detecting of object is one of the most popular topics recently, which used for motion detection of various objects on a given video or images. To achieve the goal of intelligent navigation of a moving platform operating on the sidewalk, our goal is to build the software that is able to detect the pedestrians and predict their trajectories, during which process MOT (multiple object tracking) plays an important role. Although different kinds of approaches have been introduced in the class, even in latest research papers, to tackle similar problem, there still exists many issues unsolved. In this report, we inspect the previous work and propose our method for better combination of detection and MOT algorithms. Here we take advantage of the high efficiency and celerity of Faster RCNN (Region-based Convolutional Neutral Network) and KCF (Kernelized Correlation Filter) for the purpose of realtime performance. To validate the effectiveness of our system, the algorithm is demonstrated on four video recordings from a standard dataset. The experimental results on our test sequences illustrate the reliability of our method. Keywords—Visual detection, Multiple object tracking, Neutral networks, Kalman Filter, Kernelized Correlation Filter.