Study on the Possibility of Detecting Objects in Real Time on a Mobile Robot

Abstract

Today, a task of current interest in the field of artificial intelligence in digital image processing is the detection of objects using a convolutional neural network. The purpose of this work is to study the processing of video stream in real-time with the help of a modified tracking module on the client-server system used in robotic complexes. The modified tracking module proposed in this paper, which is a combination of the KCF and SORT algorithms, eliminates object detection duplicates and levels the low speed of the convolutional neural network. By measuring the operating time of each module in the system, was obtained the frame rate of each module. The obtained time characteristics of the client-server system modules confirm the effectiveness of the proposed modified tracking module. The practical significance of the work consists of the hypothesis confirmation is about reducing the impact of the object detection rate on the overall performance of the system when using the tracking module.