抄録
We have proposed a technique to recognize a vehicle using color distribution.
Proposed technique targets the image of the vehicle viewed from rear.
A vehicle is recognized by detecting particular objects of a vehicle such as a window, tail lights and so on.
Then color distribution have been approximated by Gaussian Mixture Model(GMM) to reduce data size and to represent color distribution using the parameters.
Vehicle appearance changed by imaging condition such as time, weather and so on,
so the imaging parameters are introduced which represent imaging condition.
GMM parameters such as weighting coefficients, an average, and a variance are affected by imaging condition.
As imaging condition changes successively, GMM parameters also changes successively.
In this paper, we propose a method for training GMM parameters which reflect the successive change of imaging condition.
Then proposed method can speed up the training of GMM parameters.
Experimental results show that GMM parameters for unseen imaging condition are estimated accurately
and training are speeded up by proposed method.
