A 2D/3D Vision Based Approach Applied to Road Detection in Urban Environments

Abstract

This paper presents an approach for road detection based on image segmentation. This segmentation is resulted from merging 2D and 3D image processing data from a stereo vision system. The 2D layer returns a matrix containing pixel's clusters based on the Watershed transform. Whereas the 3D layer return labels, that are classified by the V-Disparity technique, to free spaces, obstacles and non-classified area. Thus, a feature's descriptor for each cluster is composed with features from both layers. The road pattern recognition was performed by an artificial neural network, trained to obtain a final result from this feature's descriptor. The proposed work reports real experiments carried out in a challenging urban environment to illustrate the validity and application of this approach.

Comprehensive performance analysis of road detection algorithms using the common urban Kitti-road benchmark

“…The time to train each category takes around 24 hours and the classification time for a single image takes around 3 seconds. It should be mentioned that some adaptations from the first version proposed in [11] were required: (i) it does not use the moving average technique, because it is applicable only for image sequences, (ii) the learning process does not use the strategy of training with subclass (shadow area, normal area and land marks), where it would improve the final result of road detection. In the case of the learning process to Joint Boosting algorithm, the training images were splitted into training subset containing 40%, and the test subset with 60%.…”

“…The second row presents the ground truth, followed by the next two rows that are the HistonBoost result and Artificial Neural Network result. complexity of the Kitti-road dataset, or, the training process used for ANN was not adequate, having low expressiveness if observed the strategy of subclasses used in [11]. To conclude the evaluation process, Table IV presents the final results merging all categories.…”

“…The AreaClose filter is employed to filter out connected components smaller than a given parameter λ, followed by the Hmin filter that is responsible for eliminating the local minimum given a second parameter h. The main goal of these filters is to give flexibility allowing the Watershed Transform to determine the segmentation result, i.e, ranging the parameter λ and h, it is possible to define a fine or a coarse segmentation result. This process follows the same approach proposed in [11] and its result can be seen in Figure 1a.…”

“…Texture is scale-dependent and is affected by the strong perspective in road image. Some approaches propose improvements in the model by merging color, texture and adding 3D information [9][10] [11].…”

Road detection in image by fusion laser points based on fuzzy SVM for a small ground mobile robot

“…For another hand, road detection methods can also be categorized into feature-based [15,18,19,42], [16] 2008 Laser range finder, HDL 64E Lidar, binocular camera system Urban Junior 3 [17] 2010 Laser range finder,Riegl laser, HDL 64E Lidar, Bosch Radae, colored camera Urban VRC [36] 2013 Laser range finder, binocular camera system Complex environment model-based [25,44,46] and region-based [11,19]. Feature-based method extract road features such as color, texture, boundary, etc.…”