PITCH ANGLE ESTIMATION USING A VEHICLE-MOUNTED MONOCULAR CAMERA FOR
RANGE MEASUREMENT
Li, Bo
a
, Zhang, Xiaolin
b
, Sato, Makoto
c
Department of Information Processing, Tokyo Institute of Technology, Tokyo, Japan
Shanghai Institute of Microsystem And Information Technology, Chinese Academy of sciences, Shanghai, China
Precision and Intelligence Laboratory, Tokyo Institute of Technology, Tokyo, Japan
snakeie@gmail.com, xlzhang@mail.sim.ac.cn, msato@pi.titech.ac.jp
ABSTRACT
To achieve a usable range measurement feature using a
Vehicle-Mounted monocular camera for ADAS (Advantage
Drive Assist System), a pitch angle estimation with high
accuracy is required. We propose a method for estimating
pitch angle with a non-occurrence of cumulative error, and the
initial pitch angle can be estimated simultaneously with only a
Vehicle-Mounted monocular camera. We use the Harris-
corner algorithm and the pyramid Lucas-Kanade method to
detect the optical flow of feature points between adjacent
frames from the monocular camera. With the result of the
optical flow detection using Structure from Motion method to
estimate the camera ego-motion parameters, including the
rotation matrix and the translation vector, and to optimize the
estimated ego-motion parameters using the Gauss-Newton
method. In addition, we propose a method of estimating pitch
angle relative to the road surface from the translation vector.
The pitch angle and the pitch angle rate decomposed from the
rotation matrix of the adjacent frames are composed using an
average transfer method, to achieve the high accuracy pitch
estimation. Further, the effectiveness of the proposed method
is confirmed by experiment.
Index Terms— Pitch angle estimation, pitch angle
composition, camera ego-motion estimation, range
measurement by vehicle-mounted monocular camera, average
transfer method
1. INTRODUCTION
Environmental sensing technology that can recognize
vehicles, pedestrians, and traffic signs and can measure the
range of targets constitutes the basic requirement to build
Advantage Driver Assist systems and Active-Safety Systems.
Lidar and millimeter wave radar are able to measure the
distance of targets, but the attribute of targets such as vehicles,
pedestrians or another obstacle could not be identified. As a
type of environmental sensing technology, a vehicle-mounted
monocular camera and a stereo camera with advanced image
processing can capture both the attribute information and the
distance information of the targets. In this paper, our interest is
providing safe driving for everyone; therefore, we focus on the
monocular camera that is most advantageous in terms of cost.
The identification of the attribute of targets using feature
detection algorithms, such as HoG [1], and machine learning
methods, such as ada-boost [2], based on a vehicle-mounted
monocular camera provides better performance compared to
millimeter wave radar or lidar. However, millimeter wave radar
and lidar are more stable for range measurement. To achieve a
vehicle-mounted camera based ADAS or Active-Safety
system, the range measurement feature is crucial. Even for a
data fusion system based on camera and radar, in order to
achieve the target matching function, range measurement
capability for the vehicle-mounted camera is likewise required.
Erez proposed a method [3] that could measure the TTC (Time
to collision) of the targets directly using the scale change rate
of the targets in the images for the Autonomous Emergency
Brake system and the Forward Collision Warning system, but
the range information is necessary for other functions, such as
the Adaptive Cruise Control system. A method that uses the
pedestrian’s average height or the vehicle’s average width to
estimate range was proposed, but when the height or width of
target differs markedly from the preset average value, the range
measurement performance is extremely lessened. The range
measurement using a vehicle-mounted monocular camera with
motion stereo has been attempted [4]. However, the problem
with this method is that the range of the target cannot be
accurately measured when the target is moving because the
prerequisite for using this method is that the target should be
stationary during camera movement. Using the motion stereo
method based on a reference of a stationary object that has the
same range as the moving target can resolve the problem, but
the search for the reference stationary object remains difficult.
Fig. 1 illustrates the range measurement methods using a
vehicle-mounted monocular camera.
Fig. 1. Range estimation using a monocular camera
As shown in formula (1) and Fig. 2, a range measurement
method [5] uses the depression angle of the ground point
proposed by Gideon. The problem with this method, however,
is that the change of the pitch angle due to the number and the
sitting position of passengers and the vehicle’s vibration during
driving greatly influence the accuracy of the range
measurement. Therefore, to achieve the range measurement
using the vehicle-mounted monocular camera, the estimation of
the pitch angle is necessary.
978-1-4799-2186-7/14/$31.00 ©2014 IEEE