Advance Driving Assistance Systems: Object Detection and Distance Estimation Using Deep Learning

Most of the traffic accident was caused by human error. Vehicle collision accident may happen due to the driver miscalculating the distance between other vehicles. To prevent this type of accident, we implemented an Advanced Driving Assistance System to estimate distance objects and Object detection. The architecture implemented for object detection is MobileNetV2, EfficientNet, and VGGNet16. The localization method uses Single Shot Detector (SSD). Distance Estimation method applies Depth prediction approaches using Deep Learning, with DenseDepth and MonoDepth2 as deep learning architectures. In the object detection experiment test using KITTI and PASCAL Datasets, the highest score was achieved by MobileNetV2 architecture with mean Average Precision of 75%. In terms of Deep Learning Architecture for distance estimation, comparison of prediction depth and actual distance shows that Densedepth have the lowest error with average error 3.6043 meters during the cloudy weather, and 4.0565 meters during the sunny weather.