The Vehicle Occupant Classification Based On The Multi-sensor Fusion Algorithm

With the continued increase in automobile ownership in China, the trafficaccident casualty rate is increasing, therefore, advanced automotive safety technologybecome to one of the important research direction of China’s automobile industry.Automotive safety technology consists of two parts, active safety technology andpassive safety technology. In the passive safety, developing a new adaptive airbags isneeded in order to maximize the protection of the occupant.Traditional airbags are manufactured based on the inertial motion of the mixedclass III male dummy in the normal position when collision occurs, which could notadequately protect children, short stature adults and or occupant out of position. Thus,to identify the automotive occupant characteristic, namely occupant type and occupantposition, which is one of the most important task of adaptive airbagThe automobile occupant classification system is installed by using differentsensors around the occupant seat, such as vision sensors, pressure sensors, ultrasonicsensors, crash sensors, etc. to detect the occupant characteristic information in realtime. when the collision occurs, the occupant type and position information and otherinformation is together passed to the control center of the intelligent airbag, the controlcenter disposes this information and perform the corresponding airbags protectprogram, to provides effective protection for occupant.At present, some domestic universities and research institutions are mainlyconcentrated in a single sensor for identifying the occupant characteristic, themulti-sensor fusion research on the automobile occupant characteristic is still in itsinfancy, not yet the complete algorithm to achieve the requirements of industrialization.However a single sensor is limited to its own characteristics, the identificationaccuracy of occupant characteristic is not very high. Through the analysis of theresearch the existing occupant characteristic based on single sensor, combined withultrasonic sensor and pressure sensor, to develop a set of occupant classificationalgorithm based on the multi-sensor fusion. In the paper, the main research content is as below:Firstly, by analyzing the collected data of the ultrasonic sensor to establishoccupant posture classification system based on Threshold segmentation.Secondly, by analyzing the occupant pressure distribution data, according to thepressure change information of different occupant type and position, to developoccupant classification algorithm based on BP neural network. According to theoccupant pressure distribution, average filtering is adopted to establish the occupantmeasurement space, and then, filtered pressure data were analyzed by univariatedanalysis of variance, and extract the different types of occupant at different positionsensitive pressure points, to be the occupant characteristics space. by training occupantcharacteristic information to establish occupant type classifier and occupant positionclassifier based on the BP neural network, so as to realize the classification ofoccupant types and occupant position.finally, through the study of data fusion technology, combined with the differentcentre of gravity position of occupant at different postures, to establish of the occupantcharacteristic algorithm based on fuzzy rules, the algorithm is realized to identifyeffectively different occupant types and the occupant position, to explore a set offeasible multi-sensor occupant characteristic identification system for follow-up ofintelligent occupant protection system.