Abstract—The paper describes a mobile robot application
based on monocular vision and laser that recognizes
environment objects with capacity of avoiding obstacles in real
time. Object recognition algorithm is based on mobile robot
vision combining adaptive appearance matching and Kalman
filter. The algorithm can adjust color matching threshold
adaptively to reduce the influence of brightness variations in the
scene. First, it carries out color modeling of environment objects
in the YCrCb color space. Then it detects edge features in the
image sequence by the Sobel algorithm, and thus to identify the
outlines of targets. Finally mobile robot can recognize these
objects with high empirical probability under some prior
knowledge. Kalman filter is used as our prediction module to
search objects in view window instead of the whole image
sequence and to reduce calculation time. A virtual sub-targets
based algorithm is also presented for real-time obstacles
avoidance for mobile robot. Experimental results show that
objects recognition algorithm can adapt to brightness variations
and is simple, effective, easy to imply and operates with high
efficiency. Mobile robot can also avoid obstacles smoothly in
real time while navigating in the real scene.
I. INTRODUCTION
BJECT recognition and path planning are important
processes in the research field of computer vision and
intelligent robot navigation [1-3]. While moving in an
environment, a vision system has to be able to recognize the
main objects in the scene and also to avoid obstacles in the
scene. Arbitrary objects recognition in the real scenes is very
difficult and is largely unsolved. Aiming at different
application, many approaches to objects recognition have
focused, for the most part, on using local features to classify
each image patch independently. Image patch [4] and global
image features [5, 6], such as texture features [7], color
histograms [8], structural descriptions [9], Context [10], have
been used to capture key image properties for recognizing
specific objects. But most theories are time-consuming and
require robot with powerful processing systems. For
low-level robots, these recognition techniques are not
applicable because of the constraints of execution time.
Obstacles avoidance is a classical problem in robotics and
many approaches have been proposed to solve it. Several
Manuscript received September 11, 2013. This work is supported by the
National Natural Science Foundation of China (61203332).
Rui Lin is with the School of Mechanical and Electric Engineering,
Soochow University, PR China. (e-mail: linrui@suda.edu.cn).
Maohai Li is with the School of Mechanical and Electric Engineering,
Soochow University, PR China. (
*
corresponding author, Tel:
+86-0512-67587229, e-mail: limaohai@163.com).
Lining Sun is with the School of Mechanical and Electric Engineering,
Soochow University, PR China. (e-mail: lnsun@hit.edu.cn).
traditional control methods have been applied, like Potential
Field [11] and Virtual Force Field [12]. In addition, intelligent
control methods such as fuzzy control and artificial neural
network [13] have been developed greatly. Some improved
approaches have been presented to real-time obstacles
avoidance based on dynamical and uncertain environment,
such as bubble rebound algorithm [14], PVOs (Probabilistic
Velocity Obstacles) and Occupancy Grid [15], WOAH
(Working Obstacle Avoidance Heuristic) [16].
In this paper, we focus on the problem of monocular vision
for a lower-level mobile robot operating in a dynamic world.
We propose a novel real-time objects recognition and
obstacles avoidance algorithm for mobile robotics application
based on monocular vision and laser. Objects recognition
algorithm is based on adaptive appearance matching and
Kalman filter [17]. The key technique of recognizing objects
is to identify some main objects from the surrounding
environment in image sequences. Adaptive appearance
matching includes color matching and edge feature detecting.
It is carried out to reduce influence of brightness variations
and to attain motion vector of objects relative to mobile robot.
A view window is designed by Kalman filter on the image
plane synchronously. It is applied to limit possible search
space for objects recognized in the following image
sequences.
While navigating in the working scene, mobile robot
should have real-time obstacles avoidance capacity. We
propose virtual sub-targets based method to avoid collision in
dynamical environment with stationary and moving obstacles.
Fuzzy rules [18] are applied to obtain a set of virtual
sub-targets to avoid obstacles. Strategies for formulating the
fuzzy rule sets for motion and obstacles avoidance have been
proposed. Then we apply Virtual Force Method [12] to
calculate the motion control values. Mobile robot moves
toward the virtual sub-targets directly to avoid obstacles. The
method can fulfill the requirements of recognizing objects
and avoiding obstacles simultaneously in real time.
II. O
BJECTS RECOGNITION ALGORITHM
A. YCrCb Color Matching
While mobile robot is moving in a dynamic environment,
color of objects captured by vision may change a lot because
of brightness variations in the scene. In order to reduce
influence of brightness variations, we firstly utilize light
compensating algorithm in the image pre-processing. The
algorithm can automatically calculate the compensating
coefficients. The lighting-compensating equation is
Real-time Objects Recognition and Obstacles Avoidance for Mobile
Robot
Rui Lin, Maohai Li
*
, Lining Sun
O
978-1-4799-2744-9/13/$31.00 ©2013 IEEE
Proceeding of the IEEE
International Conference on Robotics and Biomimetics (ROBIO)
Shenzhen, China, December 2013