Object Recognition in High-Resolution 360° Panoramic Images Using Spherical Grids

This paper presents an innovative method for object recognition in high-resolution <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$360^{\circ}$</tex> panoramic images using a VGGl6 convolutional neural network pre-trained on ImageNet. By employing a spherical grid, the method segments panoramic images into sub-images, enabling precise classification of objects such as people and vehicles. After classification, a reconstruction process is used to reassemble the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$360^{\circ}$</tex> image, preserving its panoramic properties. Experimental results demonstrate a 98 % accuracy in object recognition, with a mean squared error (MSE) of 0.0096. This method shows potential for object labeling and recognition in ultra-high-resolution images (SK, 8K, IlK), with applications in immersive computer vision, particularly using <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$360^{\circ}$</tex> camera technology.