Occupant Type Vision Detection In Intelligent Occupant Restraint System

Intelligent occupant restraint system uses the advanced technologies of restraint system and a control system. The inputting parameters of the control system consist of occupant type, occupant position, the application of seat belt, the information of precrash and crash severity. The control system judges the inputting parameters and sends orders to seat belt system and airbag system. The seat belt system and airbag system will take action for protecting occupants according to the orders of the control system.The occupant types are one of the most basic parameters of the control system of intelligent occupant restraint system. In order to reduce the casualties of infants, children and other occupants because of airbag, the control system must exactly judge the type of occupants while crash is happening. The occupant sensor sends a part of control parameters to micro-controller. The restraint system will take action by different type occupant according to the orders of micro-controller.The research purpose of this paper is to establish an occupant type vision detection algorithm for detecting occupant type. After perusing FMVSS208 regulation, analyzing the foreign occupant feature detection system and occupant type vision detection system, researching the technology of computer vision and the technology of pattern recognition, we establish the occupant type vision detection algorithm. We use an occupant type vision detection system to validate the algorithm. The CMOS color digital camera is a part of the system. The work in this paper can be divided into 3 parts mainly:1. The research of algorithm about occupant image processing. We make through understanding of vision theory as well as image processing knowledge and extract Legendre moment vector of the occupant edge map as the feature of occupant image. The choice of feature is based on the characteristic of each type occupant edge image and the property of orthogonal Legendre moment.First we preprocess occupant image. According to the equation V = 0 .299*R+0.587*G+0.114*B, we transform color digital occupant image into grey digital occupant image. We extract interest window from occupant image and at same time transform the larger image into the smaller image under holding original information condition. Finally we compare several various image filter algorithm and choose adaptive median filter to smooth the image. Second we extract the occupant edge map from preprocessed image. After comparing several various image edge detection algorithms we choose canny edge detection algorithm to extract occupant edge map. In order to eliminate the influence of small edges, we use method of eliminating small edge to eliminate them. The Legendre moment vector is extracted from the occupant edge map that is eliminated small edges that more easily distinguish the different type of occupant.2. Classification of Occupant type spatial. According to the research of the main recognition methods of pattern recognition, we adopt occupant type classifier based on BP neural network for Classification of Occupant type spatial.. In order to overcome the disadvantages such as network learning velocity being slow, network easily getting stuck in local minima, we use BP algorithm by adding the item of the momentum and variable step-size. In order to improve generalization of BP neural network we use method of early ending.3. The experimental confirmation of occupant type vision detection algorithm. Through investigating and analyzing we choose various parts of equipment for collecting occupant images, including digital camera, digital camera's lens and so on. We collect three types of occupant images and use one thousand and four hundred images for training neural network, sixty images for validating the algorithm. The correct detecting rate of occupant type visual detecting algorithm for validating images is 88.3percent.The innovations of this paper are setting up the primary feature description of different type occupant image based on Legendre moment and using BP neural network for adaptive pattern recognition of different type occupant.