Normal view MARC view ISBD view

Comparative study of person re-identification using deep learning approaches / Thachayane Radhakrishnan

By:

Thachayane Radhakrishnan [author.]

Material type: Text

text

Media type:

unmediated

Carrier type:

volume

ISBN:

THE0009861 (Local)

Subject(s):

Dissertation note: Bachelor of Applied Science in Data Analytics with Honours-- Universiti Malaysia Pahang – 2022 Abstract: Person Re-Identification is important in video tracking applications and is still a work in progress to date. Person Re-ID is useful in real-time criminal investigation to re-identify a criminal or accused person to overcome security measures. In this work, the concentration is on Image-to-Video Person Re-ID, which relates a certain probe image to videos in the gallery. It has important applications in tracing the position of a lost individual or criminal tracking in real-time despite all the limitations such as low resolution, occlusion, angle and pose variance, illumination change, and heterogeneous matching. There are more than 100 of methods are in use today for Person Re-ID. In this project, a few deep learning approaches will be studied and compared known as Reciprocal Attention Discriminator (READ), Attention loss paired with OSM loss and CL Centers and Temporal Knowledge Propagation (TKP). These methods will be applied to the Person Re-ID setting. The effectiveness of the proposed framework was analysed using state-of-the-art Person Re-ID approaches as mentioned above on the Motian Analysis and Re-Identification Set (MARS), a benchmark dataset. The results were visualized by using Tableau software. From the results, Attention loss paired with OSM loss and CL Centers achieved high mean Average Precision (mAP) which is 82.9% compared to READ and TKP. This study contributes significantly to the wide application of re-identification systems in realistic real-life scenarios.

Tags from this library: No tags from this library for this title. Log in to add tags.

Average rating: 0.0 (0 votes)

Holdings ( 1 )
Title notes ( 4 )

Holdings
Item type	Current library	Call number	Status	Notes	Date due	Barcode
Final Year Report	UMPLIB GAMBANG	PSM .T43 2022 r Bc (Browse shelf(Opens below))	Not for loan	CD13581		T000003179

Centre for Mathematical Sciences

Bachelor of Applied Science in Data Analytics with Honours-- Universiti Malaysia Pahang – 2022

Includes bibliographical references

Person Re-Identification is important in video tracking applications and is still a work in progress to date. Person Re-ID is useful in real-time criminal investigation to re-identify a criminal or accused person to overcome security measures. In this work, the concentration is on Image-to-Video Person Re-ID, which relates a certain probe image to videos in the gallery. It has important applications in tracing the position of a lost individual or criminal tracking in real-time despite all the limitations such as low resolution, occlusion, angle and pose variance, illumination change, and heterogeneous matching. There are more than 100 of methods are in use today for Person Re-ID. In this project, a few deep learning approaches will be studied and compared known as Reciprocal Attention Discriminator (READ), Attention loss paired with OSM loss and CL Centers and Temporal Knowledge Propagation (TKP). These methods will be applied to the Person Re-ID setting. The effectiveness of the proposed framework was analysed using state-of-the-art Person Re-ID approaches as mentioned above on the Motian Analysis and Re-Identification Set (MARS), a benchmark dataset. The results were visualized by using Tableau software. From the results, Attention loss paired with OSM loss and CL Centers achieved high mean Average Precision (mAP) which is 82.9% compared to READ and TKP. This study contributes significantly to the wide application of re-identification systems in realistic real-life scenarios.