Algorithm Study On Vision Based Vehicle Occupant Classification And Position Tracking

The integration of airbags into vehicles during the1980s has significantly improved thesafety of the occupants in collision accidents. Since airbags were designed to protect a95thpercentile and in-position adult male, it may cause fatal injuries to smaller females, childrenand occupants who are out-of-position. In recent years, in order to reduce harm to occupantsof different types and positions caused by inappropriate initiation of airbag, researcherscarried out smart airbags research. Mainly on the basis of conventional airbags, the airbag’sinitiation time and inflatable strength were decided dynamically by the real-time detection andrecognition of the occupant type and position, so as to achieve the best protection to all typesof occupants.This paper presented a vision-based detection of occupant classification algorithm. Theproposed algorithm divided the occupant types on the Co-driver seat into four categoriesincluding empty seat, adult, child and infant. The process consisted of three parts, namelygeneration of the measuring space, extraction of the feature space and classification of thetype space. Firstly, edge detection was done with prewitt operator after preprocessing and thecutting of region of interest. Then, occupant features were extracted by Legendre moments.Lastly, the Support Vector Machine (SVM) classifier was used to train the known occupants’features of different types and to determine the unknown occupant type based on therepresentative features. The smart airbag would judge and make a decision based on therecognition results. If the occupant recognized was an adult, its real-time position would betracked. The background subtraction and skin color based face recognition algorithm wereapplied to extract the occupant’s upper silhouette after preprocessing. Then, the uppersilhouette was fitted with an ellipse and the ellipse was tracked in real time. At last, the in orout of position state was judged through the ellipse equation.Test of classification with600sample occupants’ images had been done and an accuraterate of more than99%was achieved. The occupant position tracking algorithm was verifiedwith adult’s video images. The experiment results showed that the vision-based detection ofoccupant classification and position tracking algorithm could classify and track the occupantscorrectly and accurately, which provided important control parameters for the development ofsmart airbag.Some preliminary exploration and beneficial attempts had been done about the unitedprogramming using MATLAB and DSP, which provided a new way for DSP program’s fast developing and had an important significance for shorting the development cycle andimproving work efficiency.