The Study On The Algorithm Of Forward Collision Avoidance Warning System Based On Machine Vision

At the present time, a better solution to solve the problem that the high cost of the automobile’s active safety system prohibits its widespread application is to apply a single low-cost sensor in the system. So the machine vision-based anterior anti-collision technology has become the hot research in the field of active safety at home and abroad. And the data shows that only give drivers one second before collision warning happened, nearly 90% of traffic accidents can be avoided. Therefore it is important and significant for us to study the algorithm of forward collision avoidance warning system based on machine vision.In this paper, the forward collision avoidance warning system based on machine vision algorithm research is mainly to study the driving environment and parameters of vehicle and then to judge traffic safety state and establish an avoidance maneuver. So obstacle detection is the key to our system algorithm, and huge amount of useful information and little redundant information is the premise of efficient obstacles detection, therefore the image filter is the indispensable. To solve the problem that the loss of information on image edge was often found in image denoising process, this paper proposes an edge-preserving filtering algorithm based on the traditional Gaussian filtering algorithm, which combines pixel gray value differences. Finally we use charts and statistics to made sure of this method. This paper proposes an obstacle estimation method using optical flow field based on TV-1L model in order to further improve the robustness, precision and efficiency of obstacle detection. This algorithm is mainly composed of three parts: 1L norm spectral flow model, Gaussian smoothing method and non-local median filtering method. The experimental results indicate that the algorithm has strong anti-noise ability, good real-time and high precision, etc. Through the analysis of the automobile brake process analysis, the automobile safety distance calculation model based on road adhesion coefficient is set up. In this model, we consider three cases: vehicles ahead stands still, vehicles ahead moves at a constant velocity or acceleration, vehicles ahead slows down or slow to the stop. At last we estimate relative distance and the collision time according to the geometrical relation projection model and camera parameter calibration method. Moreover, this paper has presented a vehicle state judgment criterion, which combined safety distance and collision time. There are three types of the vehicle state: safety state, suggestive early-warning and emergency early-warning in this paper, then we will give the different warning strategy respectively according to the three situations.