Publication Date


Document Type


First Advisor

Ferdowsi, Hasan

Degree Name

M.S. (Master of Science)

Legacy Department

Department of Electrical Engineering


Object detection and tracking are important modules that need to be mastered for safe autonomous driving. Various detection and tracking algorithms have been developed in the past decade to detect and track objects accurately and efficiently in a variety of scenarios. However, most of them perform poorly during occlusion and are unable to reassociate the object IDs after the event of occlusion. In this thesis, a single unified algorithm is developed to detect, track, and display the trajectories of multiple objects in real-time, in the perceptive surroundings of an autonomous car. An Enhanced DeepSORT (EDS) tracker is proposed to better tackle the problem of object reassociation after occlusion. EDS tracker is compared with SORT and DeepSORT trackers to evaluate its performance based on the number of times ID switching occurs on 20 different videos involving daytime and night complex road traffic environments. EDS tracker performed 44.7% better than the DeepSORT tracker when tested on an Nvidia GTX 2070 GPU. This proposed unified algorithm uses YOLOv4 for detection and an Enhanced DeepSORT tracker for tracking objects along with displaying the trajectory information in real-time. The new algorithm incorporates changes in the way DeepSORT treats the unmatched tracks by introducing a robust method to preserve the object IDs in the event of occlusion. The importance of obtaining a birds-eye view for effective trajectory generation has also been discussed in brief.


64 pages




Northern Illinois University

Rights Statement

In Copyright

Rights Statement 2

NIU theses are protected by copyright. They may be viewed from Huskie Commons for any purpose, but reproduction or distribution in any format is prohibited without the written permission of the authors.

Media Type