Precise Detection of People Using Pre-Trained Machine Learning Models With Assisting Heat Sensor Camera
(2021) EITM01 20211Department of Electrical and Information Technology
- Abstract
- The thesis examines the possibility of developing a system that can be used to detect the number of people in a given area with high accuracy. By using a variety of hardware components, publicly available object detection models and a cloud platform a highly modern approach is examined. The aim is to develop a cost-effective and scalable system which should be able to adapt to a given area. Different modern object detection models are being evaluated based on selected metrics to optimize the outcome in the given area. The thesis is based on previous research and work in the field. The approach is based on collecting data locally using an edge device and forward relevant data to the cloud for further analysis. By enabling an interchangeable... (More)
- The thesis examines the possibility of developing a system that can be used to detect the number of people in a given area with high accuracy. By using a variety of hardware components, publicly available object detection models and a cloud platform a highly modern approach is examined. The aim is to develop a cost-effective and scalable system which should be able to adapt to a given area. Different modern object detection models are being evaluated based on selected metrics to optimize the outcome in the given area. The thesis is based on previous research and work in the field. The approach is based on collecting data locally using an edge device and forward relevant data to the cloud for further analysis. By enabling an interchangeable model for object detection, publicly available object detection models can be reused to evaluate its performance in a given area. External thermal data is used to validate a detection to achieve a more accurate detection, thereby extending the scope of the system. The proof of concept intends to demonstrate that the system described can be developed and evaluated with limited financial resources. Further analysis is intended to reveal further uses of the system, but since this is not the primary objective, it will be discussed in a purely abstract manner. (Less)
- Popular Abstract
- An increased interest in data collection and data usage in combination with a global state, the current pandemic, sets requirements for products / services that can detect the number of people in a given area. A system that helps with the ability to comply with restrictions, recommendations or other reasons is the basis for the thesis.
By using a combination of different modern techniques and publicly available material the thesis demonstrates how a system that detects the number of people in a given area with a high accuracy could be designed with a limited financial budget.
The system is designed as follows, data is collected and handled locally in a pre-defined area. Computational power demanding tasks are delegated and handled... (More) - An increased interest in data collection and data usage in combination with a global state, the current pandemic, sets requirements for products / services that can detect the number of people in a given area. A system that helps with the ability to comply with restrictions, recommendations or other reasons is the basis for the thesis.
By using a combination of different modern techniques and publicly available material the thesis demonstrates how a system that detects the number of people in a given area with a high accuracy could be designed with a limited financial budget.
The system is designed as follows, data is collected and handled locally in a pre-defined area. Computational power demanding tasks are delegated and handled externally. The information about the current condition in the given area can be requested, i.e. the number of people present in the given area. The delegation and information retrieval is handled through API calls, request and response.
The proof of concept evaluates different models for detection and examines potential benefits of using thermal data as an additional attribute. The thesis advocates further development through a system design that makes it possible to replace the object detection model for further evaluation and by proposing alternative uses for the system.
The system is developed through a combination of experience and research in the area. The system detects the number of people in an area with a high accuracy using a pre-trained object detection model on its own and shows how thermal data can be used to improve detection. Since the thesis is a proof of concept, optimization has had a low priority but with the intention of developing an optimized system with a purpose, to solve a presented problem.
The thesis focuses mainly on the technical aspects related to data collection, collaboration between components and the use of different detection models, but also takes into account some legal and ethical aspects the thesis includes. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9057133
- author
- Joelsson, Viktor LU and Rudberg, Elias LU
- supervisor
-
- William Tärneberg LU
- Maria Kihl LU
- organization
- course
- EITM01 20211
- year
- 2021
- type
- H2 - Master's Degree (Two Years)
- subject
- report number
- LU/LTH-EIT 2021-827
- language
- English
- id
- 9057133
- date added to LUP
- 2021-06-24 14:31:28
- date last changed
- 2021-06-24 14:31:28
@misc{9057133,
abstract = {{The thesis examines the possibility of developing a system that can be used to detect the number of people in a given area with high accuracy. By using a variety of hardware components, publicly available object detection models and a cloud platform a highly modern approach is examined. The aim is to develop a cost-effective and scalable system which should be able to adapt to a given area. Different modern object detection models are being evaluated based on selected metrics to optimize the outcome in the given area. The thesis is based on previous research and work in the field. The approach is based on collecting data locally using an edge device and forward relevant data to the cloud for further analysis. By enabling an interchangeable model for object detection, publicly available object detection models can be reused to evaluate its performance in a given area. External thermal data is used to validate a detection to achieve a more accurate detection, thereby extending the scope of the system. The proof of concept intends to demonstrate that the system described can be developed and evaluated with limited financial resources. Further analysis is intended to reveal further uses of the system, but since this is not the primary objective, it will be discussed in a purely abstract manner.}},
author = {{Joelsson, Viktor and Rudberg, Elias}},
language = {{eng}},
note = {{Student Paper}},
title = {{Precise Detection of People Using Pre-Trained Machine Learning Models With Assisting Heat Sensor Camera}},
year = {{2021}},
}