Research on high-definition video vehicles location and tracking Xiong Changzhen, LiLin IEEE, Distributed Computing and Applications to Business Engineering.

Research on high-definition video vehicles location and tracking

Xiong Changzhen , LiLin IEEE, Distributed Computing and Applications

to Business Engineering and Science (DCABES), 2010 Ninth International Symposium on

1 、 Introduction

• high-definition(HD) video detection system can achieve high-definition monitoring to urban roads.– Managers can analyze clearly the road traffic

condition and traffic flow vehicle character.• With the capability of intelligent analysis to

high-definition images, it can achieve road traffic flow detection and incident detection.

1 、 Introduction

• The existing vehicle detection algorithm of standard-definition(SD) video mainly includes: – Optical flow method – Frame difference method– Edge detection method– Gaussian background modeling method– Bayesian background modeling method – Background modeling based on codebook method

• this paper presents a vehicle tracking method based on image brightness.

2、 HD (high-definition) video vehicledetection algorithm

• The road can be divided into the lane area – Because interference between the adjacent lanes

is relatively small.– Taking lane as a unit.

• Lane detection can also be used to adaptively demarcate detection area.

• The video with the resolution 2592 * 1936 is used in this experiment

2.1 Interest Region

2.2 Brightness Curve

Scanning lanes by line along the vertical direction in an image, summing the brightness value in line.

2.2 Brightness Curve

• Reflect the complete information– such as lane width, vehicle size, vehicle color

(black or white), moving distance of vehicles in the image, elapsed time, etc.

1. Circle out a region in the grayscale image according to the direction of traffic flow

2. Denoise the stored values (removing the values which is too large or too small)

3. Draw the curve.

2.3 Vehicle Location

• Take the absolute value after difference with the background value. • Adaptive Gaussian background modeling method

Input image

Is initial background?

Pre-processing

-Adaptive Background production

Background

yesno

Background update

2.3 Vehicle Location

• Adaptive Background Extraction:

d>T =>mark as uncertain backgroundd<=T =>background

Background updating

others

2.3 Vehicle Location

• we can get vehicle curve, as shown in the following formula

2.3 Vehicle Location

convex part means the vehicle's location, and we can get that the number of vehicles in this frame in this lane is 2.

2.4 Vehicle Tracking

• Count the number of vehicles. The match is taking upward position in our video

• if frame i +1 has surplus vehicles in addition to them matching vehicles, then the surplus vehicles are new vehicles, number them from top to bottom

• Number rules: number frame i from top to bottom in turn, the top vehicle is 1, the following vehicle is 2,3 • • •, the new vehicle in frame i +1 is numbered in succession

• Do the same processing to another two lanes

EXPERIMENTAL RESULTS

• The video is taken byAV5105 5 Mega pixelIP-camera high-definition camera .CPU is Intel 1.83G dual-core Dell desktop

EXPERIMENTAL RESULTS