The Research of Vehicle Tracking Based on Difference Screening and Coordinate Mapping - Advances in Image and Graphics Technologies - page 316

to defined the image interest region, and then used haar classifier to identify the vehicle
in the region of interest [2]; and Gaojian Cui used the improved region matching track-
ing algorithm to detect the vehicle [3]. Although the vision method based on stereo is
more accurate in positioning the vehicle, there are still some shortcomings like high
cost, large computation and low operation speed; the other one is to study the vehicle
tracking based on monocular camera. Using a monocular camera for vehicle tracking
has been a problem for research. Domestic and foreign scholars have corresponding
researches on this field. For instance, Gideon P. Stein used the principle of perspective
geometry to do the research of front vehicle locating and speed tracking [4]; Bing-Fei
Wu used module matching method to identify vehicles on highways, urban roads and
tunnels [5]; and Benrong Wang used the shadow features of the bottom of the vehicle
and texture features of images to identify the vehicle, and then located the vehicle via
projection transformation [6]. A method based on differential screening and coordinate
mapping is proposed in this essay to identify and locate the vehicle. As a result, the
accuracy of vehicle identification has been significantly improved and the process of
vehicle locating was simplified. Moreover, this method can accurately identify, track-
ing, positioning the vehicle in a variety of environments and can provide accurate in-
formation on working conditions for an active safety equipment like ACC system.

2 Vehicle Identification Based on Goal Differential Screening

Adaboost classifier is used in this paper for vehicle identification. The basic principle
of the haar classifier training is that a number of haar features shown in figure 1 are
used to threshold the eigenvalues of the rectangular area in images. Some haar fea-
tures selected to form a weak classifier which allows, in most of the cases, to identify
the target area in the image and refuse some non-identification target area. Depending
on different haar features, more above mentioned weak classifiers are trained to be-
come strong classifiers, and those classifiers can be used to accurately identify targets.
The progress of the target identification is showed in figure 2.

Fig. 1. Haar Features Figure

Fig. 2. Process of Target Recognize