AutoNav: A Lane and Object Detection Model for Self-Driving Cars

Abstract

The area of autonomous vehicles is of huge research interest and much has been accomplished in this area. This study involves three aspects: lane detection, object detection, and autonomous driving. Lane detection and object detection has been simulated in the CARLA simulator using TensorFlow and OpenCV libraries of Python. Canny edge detection algorithm and Hough line transform are then used to detect the lane lines. For object detection, image data is collected, labeled manually, and split into test and train data. SSD MOBNET 640 × 640 is used for training the model, and about 75% precision is obtained. Autonomous driving has been implemented in the Udacity simulator using behavioral cloning, a five-layer convolutional neural network (CNN) was used as the model and the data was trained for five epochs with 20,000 steps per epoch. Live predictions are made by the trained model which are used to run the car in autonomous mode.