| Annually, about 40% of over 500 000 road traffic accidents in China occur on city road. Today, it is of great urgent to improve city traffic safety situation while putting forward of city construction process. Driver is the primary factor that triggers traffic accident in the communications system, and the most important approach to obtain traffic information is via vision channel. For this reason, we can see the need to study car driver's dynamic visual search mode on city road.Based on physical and psychological theories of driver's dynamic vision, combined related research methods and conclusions in home and abroad, the basic modes and characteristic parameters of driver's visual behavior were discussed systemically. Several indexes were determined as basic characteristic parameters of dynamic visual search mode, including fixation duration, visual angle, visual search scope, saccade amplitude and saccade velocity.Three typical city traffic environmental features, such as passage width, driving speed and traffic sign text height, were investigated by simulating tests in proving ground. Further more, the variation rules of driver's dynamic vision in the three featured conditions were analyzed.Driver's dynamic visual search mode tests with a relative large sample were performed in real city road traffic environment. One composite route and 27 typical road sections were chosen as test roads, 20 drivers which divided into two groups were taken as objects: inexperienced group has 10 object drivers with a total driving experience covering from 2 000 to 50 000 km, while experienced group has 10 object drivers with a total driving experience covering from 50 000 to 600 000 km, and the Eyelinkâ…¡which made by SR-Research company in Canada was applied in the test. Using statistical data of different characteristic parameters, driver's eye movement features and laws in city traffic environment were analyzed from the following three aspects.(1) Driver's eye movement parameter distribution features in city traffic environment were analyzed. The distribution functions of fixation duration, saccade amplitude and saccade velocity were fitted and their goodness-of-fits were testified. It was regarded that the distributions of fixation duration and saccade velocity were approximately in accord with lognormal distribution while saccade amplitude was approximately in accord with exponential distribution.(2) Using the eye movement statistical data of experienced and inexperienced drivers, the parameters'distribution difference and mean difference were compared. The significances of their differences were testified by nonparametric analysis of variance (ANOVA). It was regarded that driving experience affected fixation duration, vertical visual search scope, saccade amplitude and saccade velocity significantly.(3) By analyzing drivers'eye movement data in different road, the parameters'distribution difference and mean difference were compared. The significances of the differences were investigated with a road type one-way ANOVA and a road type vs. driving experience two-way ANOVA.A method of using dynamic cluster theory to determine driver's visual area of fixation (AOF) was presented. In this method, the analytic coordinates of fixations in visual field were clustered to obtain different AOFs. By comparing the AOFs with test video clips, the main fixation object in different areas was determined. Compared with traditional dividing method of AOF, the method presented here is more accurate, and the statistic workload is small. Based on this dividing method of AOF, the Markov chain theory was used to solve matrix of one-step transition probabilities and stationary distribution of driver's visual behavior.A method of using fuzzy theory to classify and identify visual search mode was presented, in which, the means of driver's eye movement parameters in different road sections were standardized, and fuzzy similar matrix together with fuzzy equivalent matrix was built. Referring to the driving experience of all object drivers, the optimal classification threshold value was ascertained, and the visual search mode was divided into four classes. According to the classification standard, the visual search mode identification method was determined with principle of closeness optimization.The research was sponsored by National Natural Science Foundation (50678027).
|
PDF Full Text Request