This is a personal project under way at '자율사물 캡스톤디자인'.
Recently, newly released vehicles are equipped with a driver assistance function by default. LiDAR, radar, cameras, and ultrasonic sensors are taking the form of recognizing the surrounding situation and informing the driver through sensor fusion. Because the surrounding situation is quite accurate, it significantly reduces the risk of an accident for the driver. However, since cars are long-lived products, only a few users who recently purchased cars can use this function. So I'm trying to develop a driver assistance feature that's applicable to all vehicles.
I got the idea from a black box. Using four cameras, my goal is to recognize objects in our surroundings using YOLO and implement an Around View Monitoring (AVM) system to provide real-time information about obstacles surrounding the car using UNITY.
In this project, the AVM system aims to focus on objects further away from the vehicle, which is different to the current approach used in real vehicles, which typically focuses on the area near the bottom of the car. This adjustment will allow us to detect objects at a greater distance.
The bird's-eye-view perspective will be used to calculate the distance and location between the vehicle and the detected objects. By applying the location of the center point of the bounding box detected by YOLO to the bird's-eye-view formula, we can accurately determine the object's position.
We will also utilize YOLO to set the parameters required for the bird's-eye-view perspective. To make my idea applicable to a wide range of vehicles, similar to a black box, we aim to automate the parameter tuning process. Each camera will be responsible for detecting objects of a constant size (width x length x height), and we will automatically adjust the parameters based on the differences between the size of the bounding box displayed and the actual object size.
A simulation was made using UNITY, and 4 camera data is remapped. The detected object's location was converted into world coordinates.
This concept has the potential to overcome the limitation of existing Driver Assistance features, which are often limited to specific vehicle models from particular manufacturers. Since cars have a long lifespan once purchased, the benefits of this technology can be extended to a wide range of vehicles. Previously, obstacle location was determined through sensor fusion, and while relying solely on cameras may introduce some errors, it offers a more cost-effective solution. For last, it has the potential to prevent accidents by minimizing blind spots.