Pathway estimation for vision based road following suitable for unstructured roads

Abstract

This paper presents a novel method for monocular vision based pathway estimation. In this context, the pathway means a trajectory which an autonomous vehicle should follow to avoid departure from the road. The proposed method is based on our previous work on road region segmentation, and exploits the road region extracted from monocular sequential images to estimate parameters of the pathway model. Since the proposed method doesn't require road surface marking, it is suitable for unstructured roads, while it still works for structured roads too. The proposed method is tested on real image sequences and compared with edge based techniques.

Sequential Monte-Carlo Based Road Region Segmentation Algorithm with Uniform Spatial Sampling

“…Since the sequence is composed of road images with almost homogeneous colour, the perfect segmentation of road region can be done very easily, and this sequence can be used to assess quality of segmentation. C: Real Image Sequences Real image sequences, already mentioned in report [16], with representative frames shown in Fig. 8, were prepared.…”

Improved Algorithm for Road Region Segmentation Based on Sequential Monte-Carlo Estimation

“…Real image sequences, already mentioned in our report [14], with representative frames shown in Figure 6. These sequences are real data, therefore they can be used to assess overall ability of the algorithm.…”

“…As a quantitative measure of the segmentation ability, we adopted Jaccard index J , defined by eq. (14).…”

“…SRG based algorithm, original SMC based algorithm and proposed SMC based algorithm were evaluated for data A, B, C. Before evaluation, all necessary parameters were once tuned to maximise average value of Jaccard index eq. (14). After tuning, parameter values were fixed, and same fixed values were used for all tested data.…”

“…In our work, we have focused to approach of statistical decision, and attempt to solve estimation of pdf for sequential image in more systematic way. In our previous report, we have already proposed a road segmentation algorithm based on sequential Monte-Carlo (SMC) estimation [13], and applied it to region-based pathway estimation [14]. Although this segmentation method yields better results than the conventional one, and it is quite robust to parameter setting, it still doesn't produce optimal result under certain circumstances.…”