Versatile Object Counting System (VOCS)

• Main Authors: Apolonio, Daniel Rene T., Cruz, Marc Francis G., Sia, Jackson N., Wong, Eugene W.
• Format: text
• Language: English
• Published: Animo Repository 2006
• Subjects: Counting; Algorithms; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/14186
• Institution: De La Salle University

Automated object counting is the process where a computer and other relevant hardware are utilized to track, identify and count the number of objects in a region. The purpose of automated object counting system is to provide fast and objective results with few errors. However, majority of the current automated counting system have a fixed design where the user has no choice but to adopt to how the system works. Consequently, this limits the capabilities of the system in recognizing objects and the use of the system for only a particular application. This research study entitled Versatile Object Counting System (VOCS) intends to create an automated object counting system that uses different Digital Signal Processing (DSP) techniques, separates adjacent objects, and bases its criteria on user-selected attributes in counting objects. The attributes include shape, size, color, texture and orientation. For each selected attribute, the user can vary the respective similarity measure. These features result to a flexible automated object counting system that is applicable to almost any field or real world application such as robotics, medicine, security systems, and in the industry. VOCS employs functions that perform noise filtering and object segmentation on an input image before the attributes of each individual object are extracted. Different algorithms are implemented to determine the similarity between the attributes of an input object and the reference object. Should the user-defined criteria be satisfied by the similarity, then the input object is highlighted and tallied for user verification. VOCS is able to locate, identify and count objects in the input image. Object detection is 100% accurate for input images containing similar objects and 98.89% in the case of different objects. The system is able to correctly detect an object through its boundary provided that the contrast between the object and background is high. In cases of touching objects, the system has the capability of separating these objects so that theses are not counted as one. The algorithms used for the separation of adjacent objects yields an accuracy of 92.86% for round objects, 41.9% for rectangular objects, 49.49% for irregularly shaped objects and 11.54% for long and thin objects.