Improving ground detection for unmanned vehicle systems in environmental noise scenarios

Drivable ground detection contributes to the perfect navigation of unmanned vehicle systems, and this significantly attracted the attention of researchers. Various vision-based techniques have been proposed, and amazing results are accomplished towards drivable ground detection. Despite these achievements, environmental noises like snows, rains and shadows have an effect that can lead to mis-detection of drivable ground. After conducting a brief study on snow, rain and shadow, we introduce filtering algorithms into the drivable ground detection system to overcome the effect of these environmental noises. Experimental comparison was carried out qualitatively and quantitatively. Quantitative experiments were carried out using these following schemes: accuracy rate (ACC), error rate (ERR), true positive rate (TPR), false negative rate (FNR), true negative rate (TNR), false positive rate (FPR) and precision (PRE). These schemes are used for a comparison between the system with and without a filtering algorithm. The results from all experiments show improved performance of the system with a filtering algorithm over other systems, and these massively contributed to perfect navigation for the unmanned vehicle system.