A Self-Driving Car Implementation using Computer Vision for Detection and Navigation

In recent years, lidar has been used as primary sensors for self-driving cars, however, due to their high expense, it becomes infeasible for mass production. Hence, we present the working of a self-driving car prototype that relies upon a cheaper alternative, viz. cameras. The primary objective of our prototype is to navigate safely, quickly, efficiently and comfortably through our virtual environment using computer vision. We have performed detection of lanes, traffic cars, obstacles, signals, etc. and have used the concept of stereo vision for depth calculation. Trajectory planning and steering control have also been implemented. Experimental results show that camera-based self-driving cars are viable and thus our paper can provide a foundation for all future real-world implementations.