Traffic Light Recognition In Electronic Police System

Intelligent management of urban traffic is an important part of information construction in modern cities.The electronic police system is an important application of the intelligent management of the city.It not only ensures the safety of the vehicles running and maintains the traffic order,but also improves the management efficiency and maintains the traffic order,but also improves the management efficiency and execution.As one of the key functions of the electronic police system,the identification of traffic signal has been concerned by domestic and foreign scholars.However,due to the complexity of the urban environment,there are many factors that interfere with traffic lights.For lack of existing algorithms and electronic police systems,we propose an algorithm for real-time recognition of traffic lights throughout the day.It includes color and type of identification of traffic lights.During the color detection phase of traffic lights,for the questions that the result of detection is ineffective in the existing algorithm,we propose an algorithm to detect a sub-periods of traffic lights.It includes the extraction and filtering of traffic lights candidate region;During the type recognition phase of traffic lights,for missed and false detection problem in existing algorithms,we propose a multi-frame correlation optimization algorithm,which associates the tracking results with the detection results.And then use the characteristics of cyclical changes in the status and type of traffic lights to optimize recognition results.Experimental results show that is complex urban environments the algorithm accuracy rate of more than 90%,and the processing time is 20 images per second.The algorithm is suitable for all-day,arrow type and circular traffic lights,column and suspended installation of traffic lights,so it has good accuracy,practicality and robustness.