Effective Object Detection and Tracking for Holonomic Robot using Deep Neural Architecture

The power of Deep Learning-based Computer Vision is not only making Object Detection tasks efficient but also more interesting. It gives an ability to computers for performing complex tasks as humans by locating and distinguishing between objects. The present research paper aims to create a robust tracking algorithm based on a custom-built dataset of a rugby ball, generated to replicate industrial objects. With the rapid growth in the automation industry, the need for compact yet reliable computational algorithms is a necessity. The primary objective behind the whole system is to achieve mobile compatibility with accurate object detection and tracking. The work also states the usage of the Convolutional Neural Networks (CNN) to increase the overall performance of pattern/image detection in tracking problems. A simple yet robust object tracking algorithm accompanied with object detection on Single Shot Multi-Box Detector (SSD) MobileNet architecture is applied in this paper. The method offers a minuscule model compatible to run on embedded systems without trading off much performance.