مصطفي نوروزي

Autonomous vehicles have the potential to revolutionize urban mobility by providing sustainable, safe and convenient transportation. In recent years, several companies have introduced automation as their main research area and are also investing a large amount of their financial resources in vehicle automation. This is the period when self-driving vehicles are very close to being able to transport us to our destinations without the assistance of drivers in the near future. In this research, a problem of automatic detection and route tracking by self-driving cars has been investigated. In this regard, an algorithm for detecting the road and the vehicle ahead as an obstacle based on image processing and deep learning is presented. After that, using the recognized path, a deep neural network is designed that predicts the steering wheel angle based on the images taken from the road, and the car moves at a constant speed on the path. Also, the speed bumps on the road are detected using an ultrasonic sensor and the coordinates of the speed bump are transferred to the cloud server through the Wi-Fi ESP8266 module in the Internet of Things platform, and the speed of the car is reduced to 5 km per hour. to be Simulation has been used to eva‎luate the presented algorithm. Three databases, Tusimple and KITTI, along with simulation in unity software, have been used to check the proposed method. Based on the obtained results, the average accuracy of road and obstacle detection on the simulation data is 81.9 and its duration is 0.91 seconds, which is improved compared to the work [1] with an average accuracy of 0.86 and a duration of 2 seconds. Also, the error rate of the loss function to detect the steering angle after 30 cycles is reported to be close to zero. These results confirm the possibility of using the proposed method on self-driving cars in real conditions.

خودروهاي خودران , يادگيري عميق , ابزار يولو YOLO , پردازش تصوير

Self-driving Cars , Deep Learning , YOLO tool , image processing

Mostafa Noroozi

Dr. Satar Mirzakoochaki