Traffic Lights And Digital Detection And Recognition Systems Based On Unmanned Platforms

With the development of science and technology and the development of automobile industry,the intelligent level of automobile is getting higher and higher,and the research on the vehicle auxiliary driving and unmanned system has become a hot research topic both at home and abroad.Traffic lights detection and recognition technology,as an important component of auxiliary driving and unmanned system,attracts the attention of many researchers.At present,the research on the traffic lights detection and recognition technology is mainly focused on the motor vehicle signal lights(circle lights)and the direction signal lights(arrow lights),and for the traffic signal countdown(hereinafter referred to as digit light)less.Based on the unmanned platform,the design and realization of the traffic lights and digit detection and recognition system has the function of detecting and recognizing round lights,arrow lights and digit lights at the same time,so as to provide more decision information for the unmanned vehicles.Under the natural scene,the imaging results of traffic lights in the image affected by illumination,how to accurately locate the traffic signal is the key and difficult point to the detection process.Aiming at the characteristics of traffic lights and digits,a detection method based on saliency feature and a digit segmentation method are proposed.Firstly,color,brightness and edge features are generated on the low-resolution image and merged into a saliency map.Then,the saliency regions in the image are extracted,and the noise is filtered by geometric and color features.Finally,the double-digit regions from the candidate regions is divided into two single-digit regions by projection analysis,and obtain the candidate recognition region containing only a single digit light,a circle light and an arrow light.The experiment results show that the detection method proposed in this paper has achieved an ideal result.In the stage of traffic lights and digits recognition,a two-level classification and recognition method is proposed for the features of traffic lights and digits.In this method,a special structural classifier was designed for the first level object classification,aiming for classifying the legible objects in images rapidly and reliably.And the SVM classifier based on CNN feature and HOG feature is designed for the second level object classification in order to improve the recognition rate and robustness of the whole recognition.In the first level classification based on structural features,two sets of structural features are designed for the different characteristics of digit lights,arrow lights and circle lights.Digit light recognition is based on the seven-segment code display feature,set nine sub-regions in the digit region,and generate a nine-bit binary code by statistical bright and dark conditions of sub-regions,then obtained digit result according to the code.Arrow and circle lights are recognized by using horizontal and vertical projections to describe the target structure and then classifying them with KNN classifier.The experiment result show that the two sets of structural features proposed in this paper can obtain high recognition accuracy and can be calculated fast for legible objects.In the second classification,the combination of the more robust CNN feature and HOG feature with good geometrical and optical invariance forms a feature of higher latitude.Firstly,the ip1 layer feature of the LeNet-5 neural network model are extracted as CNN features.Then,HOG features of the traffic lights and digits regions are extracted,and the CNN features are normalized and merged with the HOG feature to form the fusion feature.Finally,SVM classifiers are used to classify objects.The experiment result show that although the proposed method is more time-intensive than the CNN classification and using only the HOG feature,the recognition rate is also higher.Finally,this paper introduced the unmanned platform,applied this system to the unmanned platform and and conducted experiment.According to different weather and light colors,analyzed the detection results of traffic lights and digits.According to different types of the lights,analyzed the recognition results,and evaluated the overall performance and efficiency of the system.The system overall achieved a recall of 96.3%with a precision of 98.1%.