Object Classification on a Conveying Belt

This paper describes a system for object detection on a conveying belt. The detection algorithms are based on the permanence effect that basically consists on increasing a state variable associated to each pixel of the sensor when the moving element is located on that particular pixel point, and to decrease it otherwise. Using this permanence effect, a characteristic called LSR can be obtained (relationship between length and speed). This characteristic generally defines the moving element uniquely, making it possible to detect it on the conveying belt. The permanence effect as well as the LSR characteristic biologically inspired mechanisms and very close to the accumulation mechanisms in the synapses of the biological neurons. The described system has learning capacity, being able to modify the group of elements to be detected without necessity of modifying the system’s algorithms. This system has been tested in the laboratory using a series of medicines, obtaining very satisfactory results. To show the versatility of the used mechanisms another application of the same system is included, in this case the recognition of faces. 1. Application description This paper describes an object classification application on a conveying belt. The aim of the application is the recognition of a set of objects from the images taken by a camera located above a conveying belt. It must be stated from the very beginning that the proposed application has only been tested as a laboratory prototype. That’s why this paper presents the results obtained in a series of simulations. The final system will consists of (a) a camera, (b) a digitalization card, (c ) a personal computer, and (d) a specific card for video processing. The application consists in the classification of objects passing in front of the camera objective on a conveying belt as indicated in figure 1. The main distinguishing characteristic of this system with respect to others designed for the same aim, lies on its learning capacity. After a learning process consisting in showing a sequence of images accompanied by a signal which denominates the appearing elements, the system is capable of classifying the elements when passing under the camera at uniform speed. Figure 1. Camera above a conveying belt. In a specific application, after the learning process, the system is able to classify and count the number of medicines (see figure 3) passing in front of the camera on the conveyor. Although this example shows the classification of a set of medicines passing on the conveyor, the system would be capable of classifying different sets of objets, such as labels, parts or others. It is essential that each element has uniform speed in front of the camera. At the end of the paper an example of face recognition is introduced. The process can be carried out in real time, analyzing every pixel in each image with a simple but specific hardware. Before accomplishing the hardware, a simulation of it has been realized and all mechanisms and algorithms on which the system is based have been tested. The simulation has been realized using the working environment shown in figure 2. As it can be seen it is made up of: (1) a black and white CCD TV camera, (2) a CCIR standard black and white television monitor, (3) a commercial image digitalization board, and (4) a personal computer. Figure 2. Simulation environment. The simulation process was designed with the possibility to control and debug computational and learning processes. The application uses as input an undefined sequence of photograms (10,000 approximately). Although the future goal of the system is to be able to work in real time, the prototype described does not yet present this working capacity, although it does process the information frame by frame, allowing the input of a new frame when the previous one has been processed. The working method of this simulation can be divided into the following steps: 1) Design and assembly of the training, test and operation sequences. This stage is carried out designing the scene background and the mobile elements. Afterwards a sequence file in which each photogram is described in a register is designed. 2) Design of the assistance values. The assistance values, used during the learning stage to indicate the application which object is being shown, are generated at the same time as the sequence files.