A Deep Learning based system for fast detection of obstacles using rear-view camera under parking scenarios

Rear-end collision warning system is an important component of Advanced Driver Assistance Systems (ADAS). Main objective is to warn the host vehicle driver about the rear-end situation while host vehicle is reversing, for example, in a parking environment. A typical rear-end collision warning system uses short range Radar or ultra-sonic sensors which either have high cost or suffer from limited performance. To overcome these limitations, a deep learning based system that employs the existing rear-view camera to detect vehicles and non-vehicles (obstacles) close to the rear of the ego vehicle, is presented. To overcome the unavailability of real-world data for training deep learning models for the above scenario, synthetic data for training has been generated. The trained model is validated on examples of real-world videos. It is observed that the proposed system is capable of identifying possible stationary and moving obstacles and generate warnings well in advance for the user to take evasive actions.