Given the growth of internet-based trading on a global level, there are several expected logistic challenges regarding the optimal transportation of large volumes of merchandise. With this in mind, the application of technologies such as computer vision and industrial robotics in facing these challenges presents significant advantages regarding the speed and reliability with which palletization tasks, a critical point in the merchandise transportation chain, can be performed. This paper presents a computer vision strategy for the localization and recognition of boxes in the context of a palletization process carried out by a robotic manipulator. The system operates using a Kinect 2.0 depth camera to capture a scene and processing the resulting point cloud. Obtained results permit the simultaneous recognition of up to 15 boxes, their position in space and their size characteristics within the workspace of the robot, with an average error of approximately 3 cm.