Study And Realization For Highway Parking Identification Based On Static Features And Parallel Processing With Multi-channel Videos

It is very dangerous for the vehicle once it comes across some extraordinary trafficconditions in highway, due to the big traffic stream and high speed of vehicle,especially the parking events. With the development of information construction, thekey roads and the tunnels of expressway have been already equipped with themonitoring equipments. Unfortunately, the parking events and other accidents weredetected with artificial monitoring mode chiefly, which means a lot of resources werewasted. Meanwhile, most of the parking events detection systems were based onfront-end instruments that lead to the low finding coverage of parking events, not onlybecause the number of video detection was limited, but also for the expensiveinstruments. Hence, it is very meaningful to study the identification of parking eventsand multi-channel video processing method at the same time, which can develop thefinding coverage of parking events and make sure the safety of vehicles in the highway.For the problems above, the thesis had studied the methods of parking eventsidentification and the detection system design. Firstly, the background modeling andupdating was studied, the vehicle target extraction based on the characteristics ofspeedway. Then the paper proposed a recognition method based on static feature ofparking events. Besides, for the requirements to increase the number of video detecting,the paper proposed a time-sharing parallel processing method to build a detectionsystem.In the aspects of parking detection algorithm, at first, the thesis proposed a quickand efficient background model based on the experience value of the road pixels for thebig traffic stream and high speed of vehicle. And then, it used the backgrounddifference method to extract moving vehicle. After that, it obtained the suspiciousstationary target according to the time series characteristics. At last, the thesis presenteda parking identification method on account of lane occupancy ratio and the lawrespectively of color, which could identify some small vehicles and differentiate themfrom the local lights. After amount of experiments, the results showed that the proposedmethod can get a higher detection rate, and reduce false detection rate.In the aspects of system designing, the thesis summarized several key functionsand designed system architecture after the demand researching. After that, it focused onthe equipment management module and data processing module. According to the requirement of administration section that they want to handle multiple videos at thesame time, the thesis came up with a time-sharing parallel processing method which canprocess different videos in a detection period. After a large number of tests, the resultdemonstrated that the method could deal with more number of videos than others,which means that it can make the requirement of Highway Group met.In the end of the thesis, it realized the system and showed the details of processes.In addition, the thesis also analyzed the performance of the parking events identificationmethods. The long-term test verified the stability of system and accuracy of thealgorithm, which showed that the detection rate was higher than90%, and the falsedetection rate was lower than10%.