Development of artificial intelligence based vehicle seat recognation system

The area of business in which motor vehicles are designed, produced and marketed is called the automotive industry. Automobile manufacturing takes place in approximately 70% of this industry. There are many parts that make up the car. It consists of about 5,000 parts such as engine, wheel, metal body and doors, seat, plastic panels. The sub-industries that manufacture each part also have an important role in the sector. In this study, seat types belonging to a vehicle model in a sub-industry company that produces seats were examined. In the automotive sector, approximately 300,000 vehicles belonging to only one model are produced annually. There are a total of two seats in the front row of each vehicle, one for the driver and one for the passenger. There are six different models when the seats are differentiated according to the fabric types. In the sub-industry companies that produce seats, an average of 600,000 seats are produced for each model annually. With the development of deep learning-based computer vision technology, the need for an application has emerged for the automatic identification and stocking of the seats produced according to their models. High inventory costs, faulty production and labels are among the biggest problems of automotive companies. With Industry 4.0, smart productions and controls can be made thanks to smart machines. High cost problems are avoided with fast, high accuracy rate and low costs. Thanks to smart machines in which artificial intelligence technologies are used, smart production technologies have been developed. Products produced with high-resolution cameras, high-sensitivity sensors, automation carrier systems are faster and more accurate. In this study, it is aimed to automatically separate the stock area and pre-production and post-production seat models in an automotive supplier industry company that produces different version seats of the automobile model. In this study, artificial intelligence method was used for automatic object (seat) detection by using car seats that were previously located in production areas and stock areas. Training and test data were created according to seat types at different environments. Test and training data were created by labeling the photographs one by one according to the model type. As a result of the trainings conducted with Faster R-CNN, RetineNEt and EfficientDet models, high test results were obtained in the Tensorflow Faster R-CNN model.