Date of Publication

8-22-2022

Document Type

Master's Thesis

Degree Name

Master of Science in Electronics and Communications Engineering

Subject Categories

Electrical and Computer Engineering | Systems and Communications

College

Gokongwei College of Engineering

Department/Unit

Electronics And Communications Engg

Thesis Advisor

Edwin Sybingco

Defense Panel Chair

Argel Bandala

Defense Panel Member

Ryan Vicerra
Robert Kerwin Billones

Abstract/Summary

With the rapid growth of the world economy, the continuous advancement of urbanization, and the continuous increase in the number of urban residents, the traffic congestion has become more and more significant. To develop public transportation so as to improve urban transportation. The transportation carrying capacity of a city provides important support and guarantee for social and economic development, improves the intelligence level of public transportation is also the main line of future public transportation development.

Public transport passenger flow is an important data source for public transport management departments to optimize urban public transport routes and vehicle scheduling. How to obtain accurate public transport passenger flow data has important research value. Nowadays, the overall framework of bus passenger flow counting is divided into three parts generally: passenger object detection, trajectory tracking and classification statistics. However, in the surveillance video, there may be changes in factors such as occlusion and light, and people's clothing is similar to the video background. All of these may lead to unsatisfactory detection results.

Therefore, this research develops a passenger counting system according to deep learning. Since the research object of this research is the passengers in the bus scene, considering the real-time nature, and the passenger head area is relatively smaller than the pedestrians in the previous research, this research chooses and improves the yolov5 algorithm by adding a 4 times downsampling to add a detection head, and uses the ciou loss function to replace the previous giou loss to enhance the detection effect of small objects. Then, applies deepsort algorithm to track the detected bus passenger targets, and finally, counts the number of bus passengers getting on or off by crossing the line. The results illustrated that the improved yolov5 algorithm increases the accuracy of object detection by 5% than original one, and this proposed method can realize bus passenger flow statistics with 93.75% accuracy.

Abstract Format

html

Language

English

Format

Electronic

Keywords

Digital counters; Bus occupants

Upload Full Text

wf_yes

Embargo Period

8-22-2022

Download

Share

COinS