INS/Image Integrated Navigation System based on Deep Learning in order to Monitor the Places Traveled by Drivers

The navigation system is one of the primary and most essential parts of a positioning system in autonomous vehicles. Inertial navigation systems (INS) based on inertial sensors are among the most common navigation systems, which have received much attention due to their low cost. However, what limits the use of this navigation system is the presence of unavoidable errors caused by inertial sensors, which grow over time. Therefore, to reduce and eliminate this error, this article proposes a suitable integrated INS/image navigation aid system for utilization in a vehicle.</strong>In this article, without the intervention of the global positioning system (GPS), the proposed system gets the map of the places our car has passed through.</strong> This map can be used to monitor the places the driver travels in order to make the traffic police systems intelligent. The vehicle is equipped with a camera and inertial sensors to take pictures of the environment. This study uses the data from the inertial sensors to increase the output data's accuracy. The photos from the camera are processed by a convolutional neural network (CNN) and give us a series of geometric features. These features are in the form of a tensor of numbers that, together with the IMU data, form the inputs of the </strong>long short-term memory</strong></em> (LSTM) netwo</strong>rk. Finally, this network gives us the vehicle's location at any moment while outperforming ORB-SLAM, CL-VO, DeepVO, and VISO2_M and is 36% more accurate than DeepVO.</strong>