Design a Support Vector Machine-based Intelligent System for Vehicle Driving Safety Warning

Abstract

This paper reports the advancement of a research extension. The outcome is a device installed in a long-haul bus for daily operation. The incumbent system features the combination of lane departure warning (LDW) function and forward collision warning (FCW) function employing the support vector machine (SVM) as the classifier. LDW recognizes the environment as in daytime or in nighttime by detecting a vanishing point and applies the appropriate thresholds for daytime and nighttime to enhance the detecting rate. The algorithmic components of LDW function include image overlapping, median filter, edge-enhancement filter and hough transform, while the FCW function identifies vehicles with a feature-based approach and verifies the vehicle candidates by the appearance-based approach. In addition, we propose a new detecting scheme by motion vector (MV) estimation, where the detection doesn't rely on the whole image inside the region of interest (ROI) but on the detection range of three different ranges to concurrently secure high detecting rate and low computing power. Besides, as distance estimation is the crucial part of FCW function, we create an innovative camera calibration algorithm working with an adjustment mechanism to enhance the accuracy of the distance estimation. The combination of refined LDW and FCW functions has successfully implemented in ADI-BF561 600MHz dual core DSP-based embedded system.