PROJECT TITLE :

Real-Time Tracking Algorithm for Aerial Vehicles Using Improved Convolutional Neural Network and Transfer Learning

ABSTRACT:

A real-time tracking algorithm that makes use of an improved convolutional neural network (CNN) and transfer learning is presented in this paper as a solution to the problems caused by traditional real-time algorithms for the aerial tracking of vehicle traffic. These problems include a lack of ability to extract image features, an excessive amount of tracking time, and a low level of accuracy. First thing that this algorithm does is compare the images from the aerial vehicle with some sample images in order to calibrate the image offset. Second, the CNN parameters are initialized by building a filter set, and transfer learning is used to build a CNN pre-training model. This is followed by an evaluation of the CNN's performance. Third, in order to extract the depth features of images, a deep convolution Feature Extraction structure map is designed to be used. In the end, the real-time tracking of aerial vehicles is finished after the depth features are used to create the target vehicle motion model, calculate the similarity between the target model and candidate models, and establish the target model's similarity to candidate models. According to the findings, the accuracy of the image correction algorithm that has been proposed is as high as 92%. The algorithm produces results that are satisfactory in terms of the extraction of features and the accuracy of calculations. In addition to this, it has a low rate of overall error, the average amount of tracking time that it requires is only 22.8 seconds, and the percentage of false negatives that it produces is as low as 0.4%. As a result, the algorithm that was proposed has a significant amount of potential practical application.