Online Multi-Object Tracking in Videos Based on Features Detected by YOLO

Younis A Al-Arbo, Prof.Dr Khalil I Alsaif

doi:10.17762/turcomat.v12i6.5801

Abstract

With the rapid development of different applications that rely on multi-object detection and tracking, significant attention has been brought toward improving the performance of these methods. Recently, Artificial Neural Networks (ANNs) have shown outstanding performance in different applications, where objects detection and tracking are no exception. In this paper, we proposed a new object tracking method based on descriptors extracted using the convolutional filters of the YOLOv3 neural network. As these features are detected and processed during the detection phase, the proposed method has exploited these features to produce efficient and robust descriptors. The proposed method has shown better performance, compared to state-of-the-art methods, by producing better predictions using less computations. The evaluation results show that the proposed method has been able to process an average of 207.6 frames per second to track objects with 67.6% Multi-Object Tracking Accuracy (MOTA) and 89.1% Multi-Object Tracking Precision (MOTP).

Highlights

In recent years, different applications, such as autonomous driving, robot navigation, video surveillance and analysis, require the use of multiple object detection and tracking
The ability of Artificial Neural Networks (ANNs) to detect features that can relate each input with the required output, i.e. distinctive features, is the main advantage of these networks, compared to earlier methods that rely on hand-crafted features, such as Local Binary Patterns (LBP) [4], Haar-like features [5] and Histogram of Oriented Gradients (HOG) [6]
We propose a new multi-object tracking method, based on the features detected by the neural network employed by You Only Look Once (YOLO) detection method

Summary

Introduction

Different applications, such as autonomous driving, robot navigation, video surveillance and analysis, require the use of multiple object detection and tracking. According to this importance, significant attention has been brought toward detecting and tracking techniques that can provide accurate and rapid predictions [1, 2]. Joint Probabilistic Data Association Filter (JPDAF) is one of the DA methods that is employed in multi-object tracking, which makes the assignment decisions by considering all the possible associations in the frame, i.e. each object detected in the current frame is compared against all the objects in the previous ones Another DA method that is applied in multi-object tracking is the Multi-Hypothesis Tracking (MHT), in which the object is assigned based on multiple associations that the object can make with previous ones. The consideration of multiple paths increases the complexity of the computations, which limits the applications of this method

Methods

Results

Conclusion