Computer Vision based Auxiliary System for Computer Assembly: System Design and implementation

Abstract

This thesis proposes a solution that employs AI in assisting a human withlittle or no technical background for computer assembly in real-time with-out any form of other assistance. To achieve this goal, a state-of-the-artobject detection, namely Lighthead R-CNN is adopted as foundation. Al-terations and modifications of the algorithm are carried out to achieve theoptimal trade-o↵between accuracy loss and speed gain. In more details,it is expected to reduce the complexity of the algorithm in order to makethe solution applicable in a computer with limited computational capacity.Numerical results show that the proposed solution is 5.25% lower in termsof accuracy but almost 8 times faster compared with the competitors. Inaddition to the objective comparisons, we did subjective testing by invitingnon-technical people to assemble di↵erent PCs with assistance of the pro-posed system. The results of the testing show a successful rate of 95.99%.It is worth mentioning that the unsuccessful cases are mainly due to theignorances of the guide or the detector by human beings, rather than erroroutput of the AI based solution.