Video_based Vehicle Tracking And Detection Algorithm Research

With the rapid development of the urban science and technology, automobile transportation has brought a lot of pressure and burden on the city's traffic, vehicle detection and tracking technology in the complex environment, as an integral part of the Intelligent Transportation Systems (ITS) is the premise of vehicle behavioral analysis and judge and has a very important practical value. In addition, along with the rapid development of intelligent vehicles autonomous navigation, lane detection and tracking quickly and accurately has become particularly important.This paper focuses on the following research:Firstly, in a single target tracking area, this paper introduces a video vehicle tracking algorithm based on Kalman and particle filter. In the course of tracking the algorithm applies the Kalman and particle filtering. Through the use of Mean Shift algorithm, the Kalman filter is added to the particle filter to calibrate the vehicle running tracking so that the experiment achieves a partial linear filtering, maintaining tracking system as a whole on the non-linear and non-Gaussian, and at the same time it takes into account the local characteristics of a linear Gaussian. Experimental results show that the proposed method can be more accurate on tracking of vehicles in a complex environment.Secondly, this paper produced a multi-target tracking based on Mean Shift particle filter. The algorithm firstly modeled the complex background of vehicle videos based on Bayesian rules, followed by the use of Mean Shift particle filter for each vehicle to predict the extent possible in the next frame, if there exists only one goal, then it is the vehicle location in the next frame; otherwise, determining their corresponding target vehicles by color similarity of Mean Shift algorithm. Since the method has predicted the range before detecting, thus avoiding a global search to improve the tracking speed and at the same time accurate tracking and robust can be achieved in part of the block and cross-vehicle in the complex background.Thirdly, this paper proposed lane detection and tracking algorithm capable of handling an arbitrary curved road. Combining the fuzzy connectivity (gray and gradient features) and Vector Lane algorithms extracts the edge control points and finally fits the edge points by the use of non-uniform B-Spline interpolation (NUBS), that is, the use of control points to be considered Vertex Fitting. The experimental results show that the algorithm reduces the shadow or other noise interference taking into account the gray and gradient feature of the road; In addition, the algorithm across small breaks of the shade of a tree or other noise. At present, the above-mentioned research areas, there are still many problems remain to be further studied, such as the existence of the vehicle shadows during detection, the vehicle detection in the poor light conditions at night or rainy weather or the background modeling among the more targets, the inaccurate lane detection within the Vision area are not accurate. These are also the further research direction...