Object detection via convolutional neural network

Abstract

Machine Learning and Artificial Intelligence are starting to gain attention around the world. Companies has begun to use it to improve lives around the world. One of the famous method is known as the object detection. This report will be covering the various object detection system that uses convolutional neural networks available. This report will also be covering the process of training a new dataset not found in pre- trained model. Starting from the pre-process of collecting or generating own datasets, creating a ground truth and increasing the count of dataset to be used for training. Uncommon objects are not easy to train without some huge datasets. Sometimes, the datasets offered are not sufficient enough to fine-tune the accuracy of the model. To make it up, simple tweak of image processing techniques could be applied with discretion. For this project, the dataset was flipped across the y-axis to increase the dataset two- folds and annotated images were chosen more meticulously to ensure that it contains variety data. This variety will make the model more robust towards unforeseen image being brought forward for object detection.