Brake-Vision: A Machine Vision-Based Inference Approach of Vehicle Braking Detection for Collision Warning Oriented System

Accidents happen everywhere and at any moment, but the road is one of the most dangerous locations where accidents occur. One of the common vehicle or road accidents is the rear-end collision, this occurs when a vehicle crashes or hits a vehicle in front of it. With the growing number of rear-end collisions, various technologies or mechanisms have been created or invented with the advent of technology to avoid and prevent rear-end collisions, such as crash sensors, a collision detection system to identify or measure the distance between two cars traveling in the same path, etc. In recent years, with the help of AI emerging technologies, some studies and research suggest brake light detection for avoidance or as prevention of rear-end collision incidents. The study focused on developing a brake light detection system for the prevention or avoidance of rear-end collision accidents using deep learning with high accuracy. The study uses the YOLOv3 algorithm for training and validation of the datasets along with the Pascal VOC and LabelImg tool for annotating the datasets. Result of testing, the system ranges from 40.0553% to 84.74234% detection accuracy. This supports that the system is capable to detect brake lights to prevent rear-end collisions