ABSTRACT
This project is an attempt at developing an object detection and counting system using modern computer vision technology. The project proposes a sack detection and counting system. This project aims at reducing the human effort involved during tracking and counting of sacks for logistics management. An acceptable technique to achieve this goal is using image processing methods on warehouse camera video outputs. This chapter presents a sack counter-classifier based on a combination of different video-image processing methods including object detection.
Manual counting of sacks can be carried out, but it takes a lot of time and requires more labor. TensorFlow Object Detection API, an open-source framework for object detection-related tasks, is used for training and testing a Faster R-CNN (reinforcement convolutional neural network) model. The model was tested as fine-tuning with a data set consisting of images extracted from video footage of sack loading in a truck. This system includes preprocessing of images, extraction of features, tracking of objects, and counting using machine-learning algorithms. This project presents computer vision and machine learning techniques for sack detection, tracking, and counting.
