Traffic Objects Detection And Recognition Based On Images And Videos

Driver's fatigue,excessive speed or the suddenly change of the front of pedestrians and vehicles often make the driver have no time to avoid the occurrence of traffic accidents,excessive speed also makes the driver ignore the instructions and warning information of the traffic signs which may cause a traffic accident when a driver is driving a car.Therefore,the study on the detection and recognition of pedestrian,vehicle and traffic signs in driving environment is of great significance to the prevention of traffic accidents.This paper chooses the pedestrians,vehicles and traffic signs in the traffic environment to be the research object.After researching the research status and development trend of the general object detection,pedestrian detection,vehicle detection and traffic signs detection and recognition algorithms at home and abroad in recent years,the research is focused on proposing an improved model which can be used to detect three kinds of traffic objects based on deformable component model,and a traffic sign recognition algorithm based on the classical recognition algorithm of traffic signs detection.The main work of this paper:1.Based on the classic pedestrian detection dataset named Caltech Pedestrian Benchmark(Caltech-USA)and the authoritative traffic signs dataset named The German Traffic Sign Detection and Recognition Benchmark,a dataset is sorted out for the research.In this paper,the pedestrian and vehicle in the partial samples of the Caltech-USA are recalibrated,the new samples and the partial samples which are used to detect the traffic mark data are integrated as the detection dataset of this paper,and some data collected from the traffic signs is as the recognition dataset.2.Considering the importance of constructing a traffic object detection model,this paper studied the traditional deformable part model based on histogram oriented gradient and propose an improved deformable part model based on a kind of new feature which fuses the histogram oriented gradient,local binary pattern and the color coherence vector in the LUV color space.The robustness and accuracy of the model is improved while the time performance of the whole model is guaranteed.3.Taking into account the practical significance of the traffic signs recognition after a detection,a lightweight convolutional neutral network is presented based on the famous traffic signs recognition network named Multi-Column Deep Neutral Network.And the data which is used to train the network is expanded by the preprocess of the three kinds of image enhancement algorithm.The experimental results based on the data set show that the proposed algorithm has high detection rate,strong robustness and low time complexity.