Multimodal interior comfort evaluation via odor and vision information
•This study addresses the limited monitoring perspectives and lack of comprehensive comfort evaluation models in ride-hailing services by introducing a genetic algorithm-optimized visual-odor multimodal in-car comfort assessment system. This paper focuses on two environmental factors with ‘long-term...
Gespeichert in:
| Veröffentlicht in: | Measurement : journal of the International Measurement Confederation Jg. 253; S. 117773 |
|---|---|
| Hauptverfasser: | , , , , |
| Format: | Journal Article |
| Sprache: | Englisch |
| Veröffentlicht: |
Elsevier Ltd
01.09.2025
|
| Schlagworte: | |
| ISSN: | 0263-2241 |
| Online-Zugang: | Volltext |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Zusammenfassung: | •This study addresses the limited monitoring perspectives and lack of comprehensive comfort evaluation models in ride-hailing services by introducing a genetic algorithm-optimized visual-odor multimodal in-car comfort assessment system. This paper focuses on two environmental factors with ‘long-term effects’—unpleasant odors and cleanliness inside the vehicle—as the research subjects. It is the first to conduct research on in-car comfort monitoring and evaluation based on a multimodal approach combining odor and visual inputs. The main contributions are as follows:•To accurately quantify the typical odor levels in ride-hailing vehicles, we have designed and developed an in-vehicle odor monitoring device. This equipment utilizes an STM32 microcontroller to read real-time sensing voltages from MQ-2, MQ-9, MQ-3, and MQ-7 gas-sensitive electronic components, thereby capturing concentration information of liquefied petroleum gas, natural gas, alcohol, and cigarette smoke inside the vehicle. To precisely characterize the cleanliness of the vehicle seats, an onboard camera is employed to capture seat images, and a lightweight VGG network is used to extract cleanliness descriptive features.•To address the dimensional disparity between odor data and image data, an odor pseudo-image encoder has been proposed. This encoder maps one-dimensional odor monitoring data into a two-dimensional pseudo-color image space, achieving dimensional alignment between odor data and image data, thereby enhancing the representational capability of multimodal fusion information. Additionally, a multimodal network model structure optimized by a genetic algorithm has been constructed for the quantitative assessment of in-vehicle environmental comfort across multiple dimensions.•A smell-visual multimodal in-vehicle comfort assessment system has been developed, which employs the Jetson Nano as the onboard computing unit to collect real-time in-vehicle odor and passenger seat image detection data for multimodal comfort evaluation. The assessment results are transmitted in real-time to a mobile application we developed for passengers, providing objective data support to assist passengers in making informed vehicle choices online.
This study addresses the limited monitoring perspectives and lack of comprehensive comfort evaluation models in ride-hailing services by introducing a genetic algorithm-optimized visual-odor multimodal in-car comfort assessment system. Specifically, the system leverages on-board cameras to capture images of passenger seating arrangements, upon which a VGG-19-based cleanliness evaluation subnetwork is constructed to effectively extract and identify cleanliness attributes within the vehicle cabin. Focusing on the common odors encountered in vehicles, an in-car odor detection apparatus is designed using MQ series odor sensors and an STM32 microcontroller. Furthermore, an odor pseudo-image encoder and an air quality evaluation subnetwork, grounded on odor concentration monitoring values, are proposed to enable the extraction and recognition of vehicle interior odor characteristics. Integrating the cleanliness and odor features, this work proposes a genetic algorithm-optimized visual-odor multimodal comfort evaluation network model, facilitating a quantitative assessment of multi-dimensional in-car comfort. Moreover, an intuitive mobile app interface is developed to display real-time data and evaluation results, thereby enhancing the ride-hailing experience. Experimental results obtained in a simulated in-car setting demonstrate that, in comparison to existing methodologies, the proposed visual-odor multimodal evaluation method for in-car comfort offers superior accuracy in assessing the in-car environment. |
|---|---|
| ISSN: | 0263-2241 |
| DOI: | 10.1016/j.measurement.2025.117773 |