Research On Pedestrian-crash Warning Of The Right Blind Area For Large Vehicles

The ownership of large vehicles is recently increasing because of flourishing economic and strong industrial transport, resulting in massive accidents with severe casualties and enormous economic losses each year. As blind spots of rearview mirrors, inner wheel difference and driver inattention exist, pedestrian, bicycles and cars may be crashed by large vehicles with serious consequences. Therefore, as one of effective means to reduce the accident rates of vehicle, driving assistance technologies get more attention. Monitoring vehicle blind spot and early warning before the collision are mandatory function of the ADAS, which can automatically detect obstacles and distance, to prompt the driver and to take the auxiliary brake mode when necessary for avoiding accidance.For reducing right-turning accident of large vehicles, the reasons of which are analyzed with the blind spots of rearview mirrors and inner wheel difference, the pedestrian detection and tracking algorithm are studied, and measurement is made for ranging how far the pedestrian to vehicles. On this basis, using an effective early warning methods in order to achieve right-turing collision warning of large vehicles. The main contents are summarized as follows:With the training, testing two modules, pedestrian detection algorithm is studied. In training module, samples are collected and trained. The use of integral image and PCA dimension reduction are to accelerate computing features, obtaining pedestrian recognition classifier. In the detection module, there is a region of interest for decresing the wrong detecting rate, as well as time costed. Finally, the algorithm is verified in a practical test and the consequence is good and rational.Meanshift tracking algorithm achieves track pedestrian trajectory with the color-texture model which adds texture trait based on color characteristic. Meanwhile, the adaptive updating algorithm of the variable window solves the problem of pedestrian changeable size when moving in the tracking process. Then accurate walking track is obtained in a tracking experiment.After a brief analysis of the current ranging mode, the camera is chosen, by which the target information is complete and more suits people’s habits of perception. Taking into account of the cost and practicality, using monocular vision is to achieve distance measurement between large vehicles and pedestrian. After analysis of the imaging model, image conversion theory, the common coordinate system of the vision measurement and transform relations are for the establishment of monocular vision distance model, which is verified.There is an approach of the fuzzy pattern recognition for early warning of right-turing large vehicles. Aftere analysis of the collected the imformation of time and distance, the warning standard model is gained by using fuzzy dynamic clustering. A principle of closeness is to gain the family, which is the danger level, for identifying samples on the basis of the fuzzy pattern recognition. Experimental verification shows the model has a better robustness.