CNN-based Visual Localization for Autonomous Vehicles under Different Weather Conditions

Abstract

 A Convolutional Neural Network (CNN) was developed for autonomous localization layerby-layer in urban driving situations.  To check the positional accuracy of the CNN, RGB images are combined with depth images using the IHS method.  With an accuracy rate of 94.74%, the simulation results demonstrated the effectiveness of the suggested strategy.  The Simulation findings demonstrate the superiority of the suggested technique for different weather conditions. Autonomous vehicles (AV) are expected to improve, transform, and revolutionize ground transportation. Previous techniques are dependent on localization employing pricey inaccuracy Global Positioning sensor. Furthermore, the performance loss is caused by drifting errors of Simultaneous Localization and M apping. Regarding categorizing and analyzing texture, cameras are much more accessible and practical. This work contributes to obtaining high accuracy for AV localization and reducing errors in predicted positions. Based on the light, accurate, and robust proposed Convolutional Neural Network (CNN), it will scale down the computational complexity and shorten the training time. Considering various weather and time of day conditions such as bright sunny, hard rain noon, and wet cloudy noon with a vision-only system Red, Green, and Blue (RGB) lowcost camera sensors. To check the positional accuracy of the CNN, RGB images are combined with depth images using the IHS method. The k-M ean technique evaluates the similarity between a specific image and all street images to obtain precise coordinates. The Simulation findings demonstrate the superiority of the suggested technique for different weather conditions, which has an accuracy of up to 94.74% and a M ean Squared Error M SE in a distance of 0 meters, as opposed to [19], where the M SE in the projected position is 4.8 meters. Another indication of the proposed method's effectiveness is that it yielded reliable results when its validity was tested on images from a dataset that had not been trained.

Localization for self-driving vehicles based on deep learning networks and RGB cameras

“…A predetermined number of photos are taken by the AV. IHS is applied to RGB and depth photos to improve dataset accuracy, comparable to the procedure employed in our published paper [51], see Figures 8. This method combines RGB and depth photos to depict the environment better.…”

“…-In the our published paper [51], AV localization results were obtained using an offline CNN-based MATLAB implementation. According to the project's specifications, the CNN architecture was modified and trained on a dataset of RGB images from the CARLA simulator.…”

Localization of Strangeness for Real Time Video in Crowd Activity Using Optical Flow and Entropy

“…Every one of the frames is obtained from a sequence that involves a movement of separate people or objects [37] Up to convergence, tracking is performed starting at the lowest resolution level. The keypoints for the subsequent level are propagated from the point locations that were discovered at a level [38]. With each one of the levels, the algorithm improves the tracking in this way [39].…”

An efficient approach for automatic crack detection using deep learning

“…In Soni et al (2022), for example, a lightweight healthcare CNN model exhibits its skill in detecting prostate cancer from MRI data, demonstrating the possibility for accurate and quick medical diagnosis. In Ghintab and Hassan (2023), CNN-based visual localization is shown to be an important advancement for autonomous cars, providing robust performance under varied weather conditions. Meanwhile, Nedeljkovic and Jakovljevic (2022) pioneers the use of CNNs to construct cyber-attack detection algorithms in industrial control systems, enhancing cybersecurity safeguards.…”