| Airbags supplement the safety belt by reducing the chance that the occupant's head and neck will strike some part of the vehicle's interior. They also help reduce the risk of serious injury by distributing crash forces more evenly across the occupant's body.However, while airbags have saved an increasing number of people in crashes, they have also been causing fatalities and injuries in a severe crash situation. But airbags were designed to protect a 95th percentile adult male in a normal position. Air bags must inflate very rapidly to be effective. The airbag with the considerable force can be quite dangerous to the occupant, particularly if the occupant is the one in out-position,a slim female or a child.In recent years, in order to reduce this possible dangerous caused by airbag in wrong situation, there has been considerable attention paid to developing'smart'airbag. The size and position of automobile occupant is used to determine the safe level of force with which to deploy the airbag.At present, more and more limousine is equipped with smart airbags. According to the United States FMVSS 208, every passenger car on the market after September of 2006 must install a smart airbags system Therefore, the development of smart airbags in the future will have a broad market prospect. Accurate and efficient occupant characteristic information is the prerequisite for Smart airbags. United States FMVSS 208 requires every passenger vehicle to be equipped with an automatic airbag suppression system for protection of children on the front passenger seat which can detect the type of the occupant with weight sensor. But this smart system only can detect the type of the occupant.Occupant characteristic refers to the occupant type (i.e. infant, child, empty seat) and the position where the occupant is seated too close to the airbag for it to be safely fired. In this thesis, occupant type categories be established by qualification in accordance with the United States FMVSS 208 standard and National Standards Body Size (GB10000-88).Three types of occupants defined in this categories specification include small body occupant, the normal body occupant and big body occupant. The occupant may be injured by airbag when the occupant seats in the out-of-position (OP) zone and critic-out-of-position (COOP) zone. There are two types of position in this thesis including in-position (IP) and out-of-position (OOP), both out-of-position (OOP) and critic out-of-position (COOP) refer to out-of-position (OOP).The goal is to detect type and position of the occupant for most Occupant Characteristic Recognition system (OCRS).In addition to weight sensors, there have been numerous other technologies proposed for occupant sensing. Most are designed either for detecting type or detecting position, but are not capable of both. These technologies are capacitive sensors,pressure sensors,ultrasound,infra-red beams and computer vision etc.At present, the Body Pressure Measurement System (BPMS) is widely used to "pressure map" people sitting on automobile seats and office chairs. The information gathered provides input for improving seat designs and comfort.We found that there are correspondence with some non-linear character between seat pressure distribution and the different types and positions of occupant.This thesis makes a new attempt to recognize occupant characteristic based only on pressure distribution sensor by intelligent pattern recognition algorithm. Support vector machine (SVM) is an excellent intelligent pattern recognition algorithm. The key features of SVMs are the use of kernels, the absence of local minima, the sparseness of the solution and the capacity control obtained by optimizing the margin. Since then SVMs have been successfully applied to real-world data analysis problems, often providing improved results compared with other techniques. A clear advantage of the support vector approach is that sparse solutions to classification and regression problems are usually obtained: only a few samples are involved in the determination of the classification or regression functions. This fact facilitates the application of SVMs to problems that involve a large amount of data, such as text processing and bioinformatics tasks. In this thesis, SVM is used of occupant characteristic pattern recognition.The research purpose of this thesis is to establish an occupant characteristic recognition algorithm for detecting occupant characteristic. After perusing FMVSS208 regulation, analyzing the foreign occupant feature detection system, researching the technology of the technology of pattern recognition, we establish the occupant characteristic recognition algorithm. We have developed an occupant characteristic recognition system to validate the algorithm. The work in this thesis can be divided into 4 parts mainly:(1) The establishment of occupant measurement space. In this thesis, the diaphragm pressure sensor used in this thesis is vulnerable to external and internal interference. There are many factors that may affect the measure accuracy, including the difference between each press measure cell, quantization noise, transmission error and human factors. In this thesis, Average filter algorithm for digital image processing is used for dealing with the pressure distribution date.(2) The acquirement of the occupant characteristics space. By the use of the sample variance, pressure sensitive point is found. And then, the pressure at pressure sensitive point is used as main feature of characteristics recognition.(3) The partition of the occupant characteristics space. After researching the technology of pattern recognition, SVM is used of the partition of the occupant type and position space. Scaling the original data before applying SVM is very important, each input feature is scaled to the range [0, 1] by linearly scaling. We find good parameters by grid-search. After the best parameters are found, the whole training model is trained again to generate the final classifier.(4) Experiment and identification of the occupant characteristic recognition algorithm. We collect 25 individual occupant effective pressure distribution samples and use 650 samples for training SVM, 127 samples for validating the algorithm.Through investigating and analyzing, we choose four kernel functions for SVM, including linear,polynomial,radial basis function (RBF),sigmoid. The maximum rate of occupant type correctly identify is 68.5% with RBF kernel function. The maximum rate of occupant sitting posture correctly identify is 84.3% with Polynomial kernel function.This thesis is purely based on the innovation of the pressure distribution data. Based on the establishment of the distribution of pressure-sensitive pressure point characteristics of the different characteristics of the main characteristics of the occupant described. Support Vector Machine and achieve different occupant characteristics of pattern recognition.In the follow-up work, we should adopt the following measures to raise recognition accuracy: (1) Use better sensor equipment to improve the accuracy of sample data.(2) Acquire more test samples of different size occupants in different position.(3) Find the optimal layout of pressure sensor.(4) Find a more suitable kernel function or establish specific new kernel function for occupant characteristic recognition.The innovations of this thesis are setting up the primary feature description of different type occupant based on seat surface pressure sensitive point and using SVM for adaptive pattern recognition of different type and position's occupant. |
PDF Full Text Request