参考:
16.Identifying Good Training Data for Self-Supervised Free Space Estimation
12.Recovering free space
of indoor scenes from a single image
02.
Real time obstacle
detection in stereovision on non flat road geometry
through” v-disparity
15.
Ground Segmentation and Occupancy Grid Generation Using
Probability Fields
In a typical traffic scenario, the following challenges and problems are faced in the analysis of the 3D environments: 1) detecting road surfaces and lane areas; 2) classification of potential obstacles above road surfaces and obtaining their 3D information; 3) classification of the roadside structures such as guardrails, lamp poles, and traffic signs.
The problem of estimating free space in structured and
static environments is usually solved by exploiting properties
of certain well defined structures. Two examples of
free space estimation solutions are that of Hedau et al. [2]
and Labayrade et al. [3]. While the first exploits the box
like geometry of furniture to estimate free space in indoor
scenes from a single camera image, the second uses the planar
geometry of a road and identifiable lane markings to estimate
the free space in urban road scenes.
In unstructured
or unknown environments such as forest areas, the lack of
structure of the scene causes methods relying on static scene
properties to fail.
The slope of the road surface shown in a road image is
coupled with a lateral slope and a longitudinal slope in
terms of the road direction. Compared with the
longitudinal slope, the change in the lateral slope is small.
Accordingly, most algorithms that estimate the slope of
the road surface have typically estimated the longitudinal
slope [1,7]. This paper also deals with the estimation of
LoPORS. The estimation of the lateral profile of the road
surface remains a problem needing to be solved.
To account for the ever changing properties of free space
in unstructured scenes, it is natural to resort to learning
based systems, which usually require a training phase in
which training data representing free space is used as an
input to the learning algorithm.
Recent free space estimation approaches tackle this
problem through self-supervision, where one classifier directly
supervises input to a second classifier. The first classifier
uses data it is confident about to label parts of the environment
as free space; this data is then provided as input
to the second classifier that extends the labeling over
the whole environment. The proposed system in this paper
lies within this framework, allowing long range fully
autonomous free space estimation without relying on any
rigid assumptions such as a planar ground or bootstrapping
methods.
A variety of sensors and sensor combinations have previously
been employed for free space estimation.Apart from the being expensive and power hungry, radars offer a narrow field of view and low accuracy in lateral
direction [4]. LiDARS are also relatively expensive and big
consumers of energy. Range measurement can alternatively
be acquired using a stereo camera rig. The advantage of such
system is its relatively low cost and the added bandwidth of
information that a picture offers; however it suffers from
drawbacks such as the requirement for calibration between
the two cameras and the degradation of depth information as
the depth increase.Due to their high resolution and precision, the use of vision sensors as the primary sense to environment perception became popular over the last decade. In contrast with other active sensors such as lasers or radars, vision sensors are passive one and provide the richest information. Although being sensitive to weathe