Detecting and Reacting to the Hazards of Passengers in Car Interior of Automated Vehicles Using Human-Pose Estimation

Many cars sold today are already able to perform automated operations, and automobile manufacturers struggle to create the most performing automated vehicle. There is, however, a lot of legislation to put in place to ensure that trials of automated cars can take place because of a lack of trust on this new technology. While carmakers aim to offer their customers the safest driverless experience possible, we investigate on the safety of the passengers inside the automated vehicle, and how dangerous behaviours can be detected and avoided when a passenger experiences a VR immersion. This will be ensured by analysing the activities of the passengers in real time during the automated drive using Human pose estimation technique.

Therefore, the goal of this thesis is to embedding a human pose estimation system in the interior of a fully automated car is catching abnormal behaviours of passengers e.g. unconsciousness, and triggering a warning or an action to increase the protection of the car passengers. Furthermore, the body-pose estimation system would prevent VR HMD users from hitting the interior of the vehicle or other passengers during the immersion.

• The taxonomy of human poses regarding different non-driving-related tasks
• The model of neural network
• The design of the system reaction on detecting dangerous body poses

• Interest in automated vehicles
• Experiences in deep learning

[1]Bastian Pfleging, Maurice Rang, and Nora Broy. Investigating user needs for non-driving- related activities during automated driving. MUM’16, pages 3–7, December 2016.

[2]Bousmalis K., Silberman N., Dohan D., Erhan D., Krishnan D. (2017). Unsupervised Pixel–Level Domain Adaptation with Generative Adversarial Networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 3722-3731).

[3]L., Fridman; E., Brown D.; M., Glazer; Angell, W.; S., Dodd; B., Jenik; H., ... & Abraham;. (2017). MIT Autonomous Vehicle Technology Study: Large-Scale Deep Learning Based Analysis of Driver Behaviour and Interaction with Automation.

[4]Martin M., Popp J., Anneken M., Voit M., Stiefelhagen R.;. (2018). Body Pose and Context Information for Driver Secondary Task Detection.