Road conditions monitoring using semantic segmentation of smartphone motion sensor data

Many studies and publications have been written about the use of moving object analysis to locate a specific item or replace a lost object in video sequences. Using semantic analysis, it could be challenging to pinpoint each meaning and follow the movement of moving objects. Some machine learning algorithms have turned to the right interpretation of photos or video recordings to communicate coherently. The technique converts visual patterns and features into visual language using dense and sparse optical flow algorithms. To semantically partition smartphone motion sensor data for any video categorization, using integrated bidirectional Long Short-Term Memory layers, this paper proposes a redesigned U-Net architecture. Experiments show that the proposed technique outperforms several existing semantic segmentation algorithms using z-axis accelerometer and z-axis gyroscope properties. The video sequence's numerous moving elements are synchronised with one another to follow the scenario. Also, the objective of this work is to assess the proposed model on roadways and other moving objects using five datasets (self-made dataset and the pothole600 dataset). After looking at the map or tracking an object, the results should be given together with the diagnosis of the moving object and its synchronization with video clips. The suggested model's goals were developed using a machine learning method that combines the validity of the results with the precision of finding the necessary moving parts. Python 3.7 platforms were used to complete the project since they are user-friendly and highly efficient platforms.