Discussion of a PhD thesis in the College of Computer Science and Mathematics - Department of Computer Science
Entitled (Detecting and Tracking Multiple Objects in a Video File)

A PhD thesis was discussed at the College of Computer Science and Mathematics at the University of Mosul on Wednesday 28-9-2022, tagged (discovery and tracking of multiple objects in a video file)
For the student Younis Abbas Younis Al Arabo and under the supervision of Prof. Khalil Ibrahim Al-Saif

The thesis presented by the student dealt with the use of two intelligent methods for the purpose of building an integrated software system that helps in the process of discovering pedestrians in the frames of the video file, and then conducting the tracking currency on them.

The study dealt with the use of YOLO (You Only Look One's) algorithm to detect the object and obtain the distinctive characteristics of each object using CNN (Convolutional Neural Network) and put them in descriptors that are configured to store these features after merging them with the visual properties of the object in relation to the matching process.

Euclidean to identify the differences between the descriptor of the current object with the object in the new frame whenever the distance between them is small leads to obtaining a match. As for the second method, it is based on the use of the Faster R-CNN algorithm (Faster Region CNN) for the purpose of obtaining the detection properties of the object. The final properties were extracted from the hidden layers close to the output layer and combined with the visual properties resulting from (pre-trained convolution) at the input layer and sent to final descriptors. Concerning the congruence, the (Deepsort) algorithm was used, which apparently relied on the (Kelman filter) and the Hungarian algorithm to search for the lowest cost to perform the matching process.

The study aims to conduct a tracking process for the object by reducing the complex mathematical operations by exploiting the characteristics extracted from the discovery process and using them in the tracking process, which leads the proposed method to avoid the use of additional physical sources to conduct the tracking, where the study also aimed to "obtain accurate tracking of pedestrians by using the function An average between the last three descriptors mitigates outliers and noise, thus reducing the value of false positives and obtaining an accurate tracking process.

The discussion committee consisted of:

Prof. Dr. Tariq Ahmed Rasheed (University of Kurdistan Hawler)/ Chief
Prof. Dr. Iman Salih Sakban (University of Babylon)/ Member
Associate Prof. Dr. Safwan Omar Hasoon (University of Mosul)/ Member
Associate Prof. Dr. Yusra Faisal Mohammed (University of Mosul)/ Member
Associate Prof. Adil Sabri Issa (University of Zakho)/ Member
Prof. Dr. Khalil Ibraheem Al-saif (Alhadbaa University College)/
Member and Supervisor


Also Read