Visual Object Tracking Approach Based on Wavelet Transforms

BOURENNANE, Mohammed (2022) Visual Object Tracking Approach Based on Wavelet Transforms. Doctoral thesis, Faculté des Sciences et de la technologie.

[img] Text

Download (4MB)


In this Thesis, a new visual object tracking (VOT) approach is proposed to overcome the main challenging problem encountered within the existing approaches known as the significant appearance changes which is due mainly to the heavy occlusion and illumination variations. Indeed, the proposed approach is based on combining the deep convolutional neural networks (CNN), the histograms of oriented gradients (HOG) features, and the discrete wavelet packet transform to ensure the implementation of three ideas. Firstly, solving the problem of illumination variation by incorporating the coefficients of the image discrete wavelet packet transform instead of the image template to handle the case of images with high saturation in the input of the used CNN, whereas the inverse discrete wavelet packet transform is used at the output for extracting the CNN features. Secondly, by combining four learned correlation filters with convolutional features, the target location is deduced using multichannel correlation maps at the CNNs output. On the other side, the maximum value of the resulting maps from correlation filters with convolutional features produced by HOG feature of the image template previously obtained are calculated and which are used as an updating parameter of the correlation filters extracted from CNN and from HOG where the major aim is to ensure long-term memory of target appearance so that the target item may be recovered if tracking fails. Thirdly, to increase the performance of HOG, the coefficients of the discrete packet wavelet transform are employed instead of the image template. Finally, for the validation and the evaluation of the proposed tracking approach performance based on specific performance metrics in comparison to the state-of-the-art counterparts, extensive simulation experiments on benchmark datasets have been conducted out, such as OTB50, OTB100 , TC128 ,and UAV20. The obtained results clearly prove the validity of the proposed approach in solving the encountered problems of visual object tracking in almost the experiment cases presented in this thesis compared to other existing tracking approaches.

Item Type: Thesis (Doctoral)
Uncontrolled Keywords: Visual tracking, deep convolution neural networks, wavelet transform, HOG features.
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering
Divisions: Faculté des Sciences et de la technologie > Département de Génie Electrique
Depositing User: Mr. Mourad Kebiel
Date Deposited: 08 Dec 2022 07:51
Last Modified: 08 Dec 2022 07:51

Actions (login required)

View Item View Item