research

related publications

THERMAL INFRARED BASED HEAD TRACKING

The feasibility of a long wavelength thermal infrared based occupant posture estimation system is investigated. The basis of this method is that the thermal infrared image can be remapped to a probability of human skin temperature and subsequently segmented in this remapped form. Currently, head location and size are relied upon for occupant classification.

SKIN PIXEL PROBABILITY REMAPPING

Each pixel of the thermal infrared image on the left is remapped to the probability of human skin pixel using a Gaussian distribution centered at the optimal intensity value, derived from temperature, for human skin. The resulting image is on the right. notice how the areas containing human skin are now brighter relative to other objects in the scene.

HEAD DETECTION AND TRACKING

Using this remapped image, the best-fit ellipse -- head location -- is determined using an edge based ellipse detector. Edges are found in the skin temperature image using Sobel operators. Given a pre-computed set of ellipse templates of various sizes, angles and eccentricities, the best fit ellipse -- head position -- is found by maximizing a cost function that takes ellipse fit and skin probability into account.

POSITION CLASSIFICATION

Position classification for airbag decisions can be made if depth can be obtained from ellipse size. This requires very precise and regular head location and size decisions. The problem is currently under investigation. Alternatives to this approach of extracting depth in a one-camera thermal infrared setup are also set for future research.

REFERENCES

For references, see related publications.