Vehicle detection and tracking from videos

• Main Author: Ros Elmilia Marlan
• Other Authors: Chau Lap Pui
• Format: Final Year Project
• Language: English
• Published: 2017
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: http://hdl.handle.net/10356/71058
• Institution: Nanyang Technological University

With video surveillance systems becoming more prominent, there have been numerous research studies carried out on detection and tracking on a subject of interest. This study aims to detect and track vehicles from video using a static camera mounted on a street lamppost using algorithms implemented on MATLAB. Five videos were executed to assess the robustness and stability of the system. The program would output two video players, namely video frame and foreground mask respectively. Background subtraction using Gaussian Mixture Model was used to detect the vehicles in the videos. Similar to theory, morphological operations helped to enhance the result on the foreground mask before BLOB Analysis were carried out to detect the vehicle by encompassing it using bounding box. Kalman Filtering was used to predict the location of the vehicles in the next frame. The bounding box would indicate the track ID of the detected vehicle, together with the word, ‘Predicted’ when the track was predicted by Kalman Filtering. The program managed to detect and track all the vehicles in the videos where the environment was not complicated and when no vehicle was present in the first few frames of the videos. The limitation of the system can be seen in the video, where the system was not able to find out the background of the video as the first frame immediately displayed a group of vehicle and also a video with occlusion issues where a group of vehicles was too close to one another. This has degraded the efficiency and performance of the system. Hence, this program works best in an environment where it was able to distinguish the vehicles from the background and distinct gap can be seen from a vehicle to another in the foreground mask, to be detected as one vehicle.