Road Segmentation for Classification of Road Weather Conditions
Using vehicle cameras to automatically assess road weather conditions requires that the road surface first be identified and segmented from the imagery. This is a challenging problem for uncalibrated cameras such as removable dash cams or cell phone camer
- PDF / 2,624,920 Bytes
- 13 Pages / 439.37 x 666.142 pts Page_size
- 57 Downloads / 206 Views
Abstract. Using vehicle cameras to automatically assess road weather conditions requires that the road surface first be identified and segmented from the imagery. This is a challenging problem for uncalibrated cameras such as removable dash cams or cell phone cameras, where the location of the road in the image may vary considerably from image to image. Here we show that combining a spatial prior with vanishing point and horizon estimators can generate improved road surface segmentation and consequently better road weather classification performance. The resulting system attains an accuracy of 86 % for binary classification (bare vs. snow/ice-covered) and 80 % for 3 classes (dry vs. wet vs. snow/icecovered) on a challenging dataset. Keywords: Linear perspective · Vanishing point mentation · Weather conditions
1
· Horizon · Road seg-
Introduction
Automatic assessment of road weather conditions using vehicle camera data can be used to inform the human driver, driver-assist controls and autonomous control systems. Moreover, the information can be shared across connected vehicles, alerting following vehicles to conditions ahead. Another application is automatic dispatch and verification of snow ploughs and service vehicles. Given their typically wide geographic distribution, these service vehicles can provide real-time data on road conditions to central management, which can then use the data to verify maintenance and optimize dispatch. While future generations of service vehicles may be manufactured with appropriate built-in cameras, in the meantime there is interest in retrofitting existing vehicles with removable dash cams that can be used for multiple purposes. This poses a challenge for video analytics, as the pose of the camera relative to the road surface may vary considerably. Since the cameras are mounted inside the vehicle, imagery may be partially occluded by the hood of the vehicle, For snow ploughs, the road surface may also be occluded further by the plough, depending on its position (Fig. 1). To address these challenges a reliable algorithm for segmenting the road surface from the imagery is required. This method must be able to handle variations c Springer International Publishing Switzerland 2016 G. Hua and H. J´ egou (Eds.): ECCV 2016 Workshops, Part I, LNCS 9913, pp. 96–108, 2016. DOI: 10.1007/978-3-319-46604-0 7
Road Segmentation for Classification of Road Weather Conditions Bare Bare dry n = 32
Bare wet n = 14
Iced covered n=8
Covered Snow covered n = 16
97
Snow packed n = 30
Fig. 1. Example images from training dataset, and the number n in each class.
in the position and pose of the camera as well as geometry of the road surface. Using appearance features of the road surface (e.g., texture) for segmentation is unlikely to be reliable across diverse weather conditions, since the road appearance will vary considerably and sometimes may strongly resemble other surfaces in the scene. For these reasons, we focus here on geometric methods for identifying the road surface and show that by fusing
Data Loading...
