Biometric measurement of one-humped camels using machine vision technology

Measuring the body dimensions in livestock is usually useful to predict the weight, grade and body score of animals. The shoulder height, chest circumference, chest depth, body length, forehead bone size, rump height, distance between eyes, ear length, and tail length are the most important body linear measurements in the livestock. Most of these dimensions are related to the live weight and some important traits of domestic animals. For example, the results of many studies indicated that the chest circumference, body length, pelvic width and shoulder height are the most appropriate and reliable parameters for estimating live weight of the animals. In recent studies, features of digital images have been used, in certain circumstances, to estimate body dimensions of domestic animals. The base of these measurements is the machine learning technology, and currently was tested on some livestock, such as dairy cattle. Therefore, this research was conducted to investigate the possibility of using machine vision technology in order to estimate body dimensions of one-humped camel.

This research was conducted on one-humped camel in a privative camel breeding herd at Qom province. The studied herd originally consisted of 9 adult mature camels, one an adult luck and 11 pedigree camels from 2 to 12 months old of age. In the following months, five baby camels were born in the herd. The body dimensions of all camels were monthly recorded, and in the same time, digital images were captured from camels regarding a constant distance (2 meters). In this study, a total of 203 biometric records of camels at different ages were measured. Each photo was first transferred to a computer, and some edits were made to improve it's quality. Twenty two morphological features from each image were extracted using defined functions of graphical user interface of MATLAB. The characteristics that were more relevant to the biometric measurement of camels were selected using Pearson correlation coefficient by SPSS software. The data mining process with the aim of discovering mathematical relationship between extracted features of digital images and body dimensions of camels was done using a feed forward neural network which was trained by the "back propagation algorithm" in MATLAB software.

Some extracted features including equivalent diameter, major axis length, minor axis length, bounding box, convex area, filled area, area ,perimeter and the number of non zero points in digital images had high and significant correlation (P<0.01) with body dimensions of camels. These features were used as effective input to design the artificial neural network. Accuracy of the artificial neural network models to estimate body length, shoulder height, and hip height of one-humped camels were 0.98, 0.96 and 0.96, respectively.

The use of image processing and artificial neural network or other data mining tools can be considered as an appropriate and accurate alternative to human assessments, and help to save the time and expense associated with the biometry of large livestock, especially camels.