Longitudinal Collision Avoidance In Self-Driving Vehicles Using Deep Learning And Particle Swarm Optimization

Abstract

Fast and efficient object detection and collision avoidance is an increasingly significant task for autonomous driving technology. This paper proposes a deep learning and swarm intelligence based approach in the automotive domain to detect objects and subsequently avoid collisions. By combining them, improvement can be achieved in the speed and accuracy of self-driving cars to avoid longitudinal collisions. Our proposed approach uses a highly accurate and well-suited deep learning technique for object detection to detect objects in real-time using algorithms and methods such as Mask Region-Based Convolutional Neural Networks (Mask R-CNN) and different versions of You Only Look Once (YOLO). Particle Swarm Optimization (PSO) is used to optimize and predict the parameters (velocity and acceleration) required for the self driving car to avoid colliding with the detected object.