The Use of Thermal Cameras for Pedestrian Detection

Visible-range camera sensors have been widely used for pedestrian detection. However, most of the methods, which employ visible-range color cameras, do not perform well under low-light and no-light conditions, e.g. during night time. Since the working principle of thermal camera sensors is mainly ba...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:IEEE sensors journal Ročník 22; číslo 12; s. 1
Hlavní autoři: Altay, Fatih, Velipasalar, Senem
Médium: Journal Article
Jazyk:angličtina
Vydáno: New York IEEE 15.06.2022
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Témata:
Cameras
Concrete blocks
Datasets
Feature extraction
Image sensors
Infrared radiation
Object detection
Object recognition
OTHER INSTRUMENTATION
Pedestrian Detection
Pedestrians
Performance evaluation
Salience
Saliency Maps
Sensors
Temperature measurement
Temperature sensors
Thermal Cameras
Thermal energy
Thermal Images
Training
Weather
ISSN:1530-437X, 1558-1748
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Visible-range camera sensors have been widely used for pedestrian detection. However, most of the methods, which employ visible-range color cameras, do not perform well under low-light and no-light conditions, e.g. during night time. Since the working principle of thermal camera sensors is mainly based on temperature and not light, they have been employed for person detection to overcome the drawbacks of visible-range sensors under these conditions. Every object gives off thermal energy, which is captured by a thermal camera sensor. When an object becomes hotter, it emits more thermal energy, and is therefore captured as much brighter or vice versa. Yet, compared to visible-range cameras, there are many additional challenges that need to be addressed when detecting pedestrians from thermal camera images. These challenges include bright hot objects close to humans, similar pixel values in an image due to weather conditions, or objects that block thermal cameras such as concrete or glass. Glass acts like a mirror for infrared radiation and reflects whatever is in front of the camera. Thus, novel methods are still required to accomplish pedestrian detection task from thermal camera images. To contribute to these efforts, we propose a new method and a modified object detection network incorporating saliency maps of thermal camera images. The features obtained from thermal images and their corresponding saliency maps are combined to obtain richer representations of pedestrian regions, and better detection performance. We perform extensive evaluations on five different datasets to compare the performance of the proposed approach with two baselines. Moreover, we evaluate and compare the transferability of these approaches by doing leave-one-out cross validation across different datasets. The results show that the proposed approach outperforms the baselines, and has better transferability properties across different thermal image datasets.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
USDOE
National Science Foundation (NSF)
AR0000940
ISSN:1530-437X
1558-1748
DOI:10.1109/JSEN.2022.3172386