Abstarct: Main objective of this thesis is to recognize predefined objects in an image of scene, especially for robotics applications. As working with robots, it is important to have a real-time algorithm, so the proposed method must have high performance in both aspects: speed and accuracy. Therefore, common methods of feature extraction (the first step in object recognition) were studied, and the SURF descxriptor, which has high accuracy and speed, has been selected. Since the extracted features are good discriminators, we used a simple method with low computational complexity to implement “matching” stage. For matching between descxriptors of scene image and object’s model we use Euclidean Distance. In this stage some outlier data appear, that were removed using RANSAC method. To implement mentioned algorithm on the robot, we must use a portable hardware; so we choose the DM6446 model in DaVinci series of TI DSP's. In this thesis, we focus mainly on setting up this Evolution Module, and also on providing a general comprehensible approach to implement arbitrary algorithms on this board.