Research On Traffic Flow Detection System Based On Video Image

Intelligent transport system(ITS) is an important application of intelligent video surveillancesystem, which will be the future tendency of this area. For the very important part of the ITS, thevision-based approach becomes popular in owing to its easy fixing, low expense and highflexibility. Most of the tradition video detection is based on the PC platform, but this methodbrings a lot of inconvenience when used in traffic scene. With the development of the computervision technology and DSP technology, a vision-based embedded vehicle detection system hasbecome possible. This kind of system overcome the defects of the traditional video surveillancesystem, because of its low power consumption, small size and flexible installation,etc.this systemhas a broad prospect in practical applications.The paper firstly studies the key technologies of domestic and international video trafficparameter extraction and extracts traffic parameter use Based on the extraction and trackingalgorithm of moving vehicles. Then the algorithm is ported to SEED-VPM642video imageprocessing platform and optimized in order to build an embedded video traffic flow detectionsystem. This paper mainly covers the following topics:1. The article compares the commonly used methods of moving vehicle detection. using theimproved surendra algorithm extract the background and detect moving regions by backgroundsubtraction method. In order to achieve accurate detection and extraction of the vehicle, the paperresearches the adaptive threshold, shadow detection and removal and connectivity analysis, etc.2. For vehicle tracking and traffic parameter extraction, the paper overviewed the commonlyused tracking methods, and then introduced the KALMAN filter principle and its application inobject tracking. In the phase of the moving object tracking, this paper presents a hybrid approachbased on KALMAN forecast and histogram matching. At last, the system completes the detectionof traffic parameters based on vehicle tracking.3. For DSP realization, the paper firstly describes the system's hardware platform andsoftware development environment. During software design, the soft framework based on TI'sDSP/BIOS multi-task operation system and FR5reference framework. The programs of allmodules are completed with C language, and then ported the algorithms to DM642kernel. Inorder to meet the real-time demand, this paper gives a series of optimization strategies accordingto the DSP hardware features.The simulation results and data from DSP show that the traffic flow parameters can beexactly and timely detected by the system under the specified occasion,it achieves the expectedeffect.