In the 2D non-contacted body measurement, the transform model which converts the human body 2D girth data to the
3D girth data is required. However, the integrate model is hardly to be obtained for the different human body type
categories determine the different model parameter. So, the work of human body type accuracy classification based on
the measure data is very important. The canonical transformation method is used to strengthen the similar of data
features of the same type and broaden the diversity of the data features of the different type. The "accumulating dead
bodies" ant colony algorithm is improved in the paper in the way of employing the road information densities to help the
ant to select the probable path lead to site of the accumulating dead bodies when it moves the data. By the way, the
randomness and blindness of the ants' walking are eliminated, and the speed of the algorithm convergence is improved.
For avoiding the unevenness of the data unit visited times in the algorithm, the access mechanism of the union data is
employed, which avoid the algorithm to get into the local foul trap. The clustering validity function is selected to verify
the clustering result of the human measure data. The experiment results indicate the affectivity and efficiency of the
human body clustering work based on the improved ant colony algorithm. Basing the sorting result, the accuracy 3D
body data transforming model can be founded, which should improve the accuracy of the non-contacted body measurement.