Driver's Intention Recognition on Roads without Crossings Using Sectional Hidden Markov Model

Abstract

In this paper, the driver's intention recognition on roads without crossings is investigated. Because of uncertain characters between driver’s intention and behaviors, analyzing and modeling methods based on the hidden Markov model (HMM) were proposed and used for setting hidden or observed states, obtaining probabilities of intention transition and observation, and building a graphical network of those definitions. In order to meet the dynamic requirement of the state transition, the structure of the HMM is improved to a sectional form, which is named SHMM, with road subsections. The transition probability of alternative intentions is designed as a time-domain continuous function, and the probabilities of observation are counted and made into statistics with the respective distribution in the subsections. The Viterbi algorithm is used for recognizing intention, with sequences of operation in the whole process of driving. The judging rule is drawn up to decide all operations in the sequence they’re driven by, i.e. which type of intention with number comparison. The data used for modeling and verifying in this study is collected by a real vehicle driving experimental platform, and the test is held on urban traffic roads with ten participants involved. The results of intention recognition under the proposed method is quite satisfying and credible.