Mendeley  

Zotero  

Mirfasihi S P, Ghazali S, Baradaran Shokouhi S. Designing and Implementing of Real-Time Intelligent System with the Ability to Identify and Classify Different Topics in Autonomous Vehicle. sjis 2022; 4 (4) :1-7
URL: http://sjis.srpub.org/article-5-172-en.html

Deep learning is a branch of the machine learning and artificial intelligence and a set of algorithms, which tries to model high-level abstract concepts using learning at different levels and layers. Due to high accuracy and efficiency, these algorithms have been used in self-driving cars. Many car accidents are caused by human errors. One of the reasons for paying attention to self-driving cars is to prevent these accidents by eliminating or reducing human interference and consequently reducing the loss of life and property. Also, save and efficient use of time is another factor that has made self-driving cars one of the topics of interest. One of the things that happen in self-driving cars is collecting information, forecasting and making decisions based on available data. In this article, we represent, review, compare and implement networks based on SSD and Mask RCNN models with two types of Inceptions and Resnet architecture in the form of TensorFlow, which is an open-source library. At the end, we have provided a table comparing these two models and representing their characteristics. To do this, we have used the Coco dataset, which is one of the most popular databases for using in mobile robotics as well as automatic driving. The SSD model with the inception v2 architecture has faster speed in identifying objects, which reaches 2fps, but in the other model, this amount is 25% less. Also, the Mask RCNN model is more accurate than the SSD model, which the observations and the results show that with more than 88% probability, it predicts the desired object correctly. All calculations have been tested on the Nvidia P1000 Quadro graphics card.