TY  - JOUR
T1  - Design and Implementation of Vehicle Operating Status Recognition On-Device AI for Driver Behavior Analysis
AU  - Kim, Taegu 
AU  - Cho, Yonghun 
AU  - Baek, Yunju 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2023
DA  - 2023/1/14
DO  - 10.7840/kics.2023.48.7.842
KW  - On-device AI
KW  - Vehicle operating status
KW  - Driver behavior analysis
KW  - Model lightweight
AB  - With the recent development of technologies for vehicle sensors and artificial intelligence, technologies for driver convenience such as autonomous driving are actively developed around the world. However, due to the verification of the safety of the system, the commercialization rate is lower than the development situation.
Therefore, in this paper, a study was conducted to classify the vehicle operating status so that it can be used to analyze driver behavior and recognize dangerous driving by implementing on-device AI available in the vehicle driven by the driver. Deep learning model was designed to infer the vehicle's operating status using the extracted vehicle interior information. In order to mount a deep learning model on a device, the structure of the deep learning model was changed and lightened through quantization. The performance is evaluated by performing real-time vehicle operation status inference while driving the finally implemented on-device AI in the real vehicle. The vehicle operating status recognition accuracy showed 91.66% performance and the inference time was 19.72 ms