Research On The Vehicle Occupant Classification And Recognition System Based On Machine Vision

In recent years, the wide use of automobile airbag greatly has reduced the casualty rate when the traffic accident happened. But ordinary airbag system is developed according to 95 percentile male occupant normal sitting positions. When an accident occurs, small stature or out-of-position passengers most likely get a secondary victimization by the airbags. Therefore, the study on intelligent airbag system will come into being.Smart airbags dynamicly decide its balloon moments, the strength of the balloon and so on, through getting the passenger’s information such as the occupant type. Then complete the best protection for different types of passengers. Therefore, the premise and the core of smart airbag system are the identification and the classification of passengers.According to the difference of various types of crew, this paper uses image processing techniques and method of pattern recognition to study the image feature extraction and selection, and the design issues of classification system. On this basis, we establish occupant classification system. The main research contents of this paper are as follows:1. The image collection and edge detection of crew. Collect the image of the passenger in the co-pilot seat by CCD camera, and then preprocess the image. Based on the crew’s image and the background inside the car, we firstly change color image to grays image, and then get a crew edge image through the image acquisition of the region of interest, image enhancement, and edge detection.2.Extract and select the feature values of the edge of images. Because the edge of the image can be seen as a random probability distribution function, so we can use torque technology to extract image feature. In this paper, Hu torque and Zernike torque are used as invariant torques of extraction of crew’s image. Characteristic vectors’ dimension not only increases complexity of calculation, and reduces the accuracy of classification. Therefore, it is necessary to optimize the extracted features. Based on search strategies and evaluation criteria, we select Differential Evolution algorithm(DE) to optimize the characteristics and combine the packaging principle. Finally, we get the best feature subset.3.Classify and identify different types of the crew. Firstly, this paper uses Support Vector Machine Pattern recognition method to study known types of crew, and the crew type recognizer is established. Secondly, verify the unknown occupant type, and get the classification’s accuracy. Compare Hu torque, Zernike torque, Hu-Zernike torque and DE selection feature vector of passenger’s classification accuracy. Experiments indicate that the eigenvector’s accuracy by DE selection is higher than the accuracy of the original feature set.