Embedded Implementation of MOD Algorithm with Image Fail-Safe for Autonomous Vehicles

Abstract

This paper aims to implement a fail-safe function based on the MOD (Moving Object Detection) algorithm on an embedded board. When the input image quality of the camera deteriorates due to a change in the driving environment, it is necessary to automatically detect the input image quality and provide a warning that the MOD function cannot be operated normally. When the vehicle equipped with a camera without the auto expose function drives in a dark area, the threshold value is determined by analyzing the brightness value of the input image, and a Laplacian filter based method is used to evaluate the occurrence of fail due to image distortion in the rain. In addition, we propose a weight reduction method for implementing the Moving Object Detection (MOD) algorithm executed in a PC environment as an embedded environment. The performance of the proposed fail-safe method was evaluated using the confusion matrix. The obtained precision and recall rates were 94.79% and 95.35%, respectively. The MOD algorithm compares the GT (Ground Truth) recorded in 15,216 images obtained in an actual vehicle driving environment and the performance realized in the PC environment by building the stored image database and carrying out the performance evaluation. In the embedded environment, the performance is realized at an average speed of 15 FPS or greater. This proved the applicability of the proposed method for realistic car operation.