Study On Intelligent Headlight Control Strategy Based On Vision

With the increase of automobile industry and car ownership, it brings serious trafficproblem and also has cased a mass of traffic accidents. The driving safety has become issuethat people concern, so all kinds of the active safety equipments were produced. Driving atnight is usually more dangerous than driving during the day. Pedestrians and cyclists on theroads are especially at high risk due to the limited visibility of motorists at night. This raisesthe importance of maximizing a driver’s forward vision for night-time driving safety purpose.One way to achieve this is to improving the utilization of the vehicle’s high-beam headlight.Nevertheless,it is dangerous and illegal to keep high-beam headlight on for ages,so it issignificant to develop intelligent headlight control (IHC) systems.This paper raises intelligent headlight control strategy based on machine vision andprovides a more detailed reference for intelligent headlight control system later. We usecamera to get the road information firstly, after deal image in the video with algorithm.When we get the ideal image, we conduct feature extraction based on vehicular feature atnight and extract feature vector of headlight (HL), taillight (TL), streetlight (SL) and other.We import the feature vector into support vector machine(SVM) to train and verify themodel using training set and test set. After recognizing all blobs within each frame using thepretrained SVM model, we elaborate on the proposed decision making mechanism. We setperformance metrics to evaluate system performance finally. The main research contents areas follows：1. Get the road information and image processing. Collecting image is the premise ofintelligent headlight control. The camera is mounted on the windshield right behind the rearmirror and faces forward. We extract the single frame image and process it. We get the imagegray scale firstly. Median filter on the image later, and we set a rgion of Interest and filterwith unsharp operator to enhance taillight in that rgion. Finally, we set threshold to do imagesegmentation.2. Feature extraction and SVM model. Finding the representative image in the videoand getting the binary image using image processing. We extract the feature vector accordingto the application and import the vector into SVM model. To prepare for the classification ofthe fllow-up work.3. Beam decision making. After recognizing all blobs within each frame using thepretrained SVM model, our last step is to make the beam decision. we assume that eachframe f has two beam states, the hidden state (HS) and the actual display state(DS). Thedifference between these two states is that, HS is determined based on what is happening inthe current frame, while DS is determined based on not only the current moment, but also thepast. Eliminating the influence of the beam frequently change and blob recognition errors. Every state have their control rules and supplement each other. Making the control strategymore perfenct.4. Performance evaluation. After accomplishing the research of control strategy. We setmany performance metrics to measure the performance. We get the video in different roadsegment and road condition and compare system results with actual situation. The systemperformance is measured in terms of false negative value (FNV), false positive value (FPV)and overall errors. We analyzes the reason of the numerical in various circumstances.Based on MATLAB software, and combined with image processing algorithm andSVM model, and put forward intelligent headlight control strategy, We set a series ofperformance metrics to test and analyze system performance using the data collected by theexperiment at last. The experimental results showed that the system has high accuracy andprovide a good reference for intelligent headlight control system.