Nighttime Vehicle Detection And Tracking

With the increasing number of vehicles recently,traffic safety issues have been taken more and more seriously.The vehicle detection in Intelligent Traffic Monitoring System has been the hot issues of concern,researches of vehicle detection system at the scene during the daytime is relatively mature,while these at nighttime scene,of which the scenes are more complex and dark,with these problems,there are difficulties in detecting cars in the nighttime,such as the reflected lights on the road would significantly influence the detection rate of headlights;while the detection methods commonly used in the daytime cannot be applied at the night scene.For these problems exist,on the premise of research of the existing algorithms at home and abroad,with the highway traffic at night as application background,we propose the corresponding vehicle detection algorithm.The main work of this article is divided into two aspects as follows:The first part of the work is based on the difference of gradient between the headlights and the reflected lights;we propose to use different classification methods in machine learning and to compare the classification effectiveness between the headlights and reflected light.First,the threshold segmentation method is based on to process the region of interest and to preliminary filter out bright prospects of the region;Then,using the principle of filtering and the Bouguer's law,according to the differences in the gradient of the headlights and reflection regions,the corresponding statistics features are built;and finally,we utilize the common classification methods in pattern recognition: Mahalanobis distance,K nearest neighbor,classification based on Bayesian decision theory,decision tree learning,the corresponding statistical characteristics are classified to eliminate reflection interference,and to achieve accurate detection of headlights.About the second part of the work,on the basis of the classification methods of the study in first part,we further explore a new method decision learning method as classification method and research a better feature which can better describe the lamps.We have found that the vehicle headlights and the reflected light have greater difference in variance feature,which means in this scenario this characteristic can be used to distinguish the headlights and reflected lights regions.So accordingly,we propose a method based on variance to achieve the night vehicle detection,and the experimental results show that on the basis of the previous work,this part of the work has achieved a more accurate and effective detection of headlights in the nighttime.The third part of the work is mainly about the vehicle tracking,based on the research and analysis of existing tracking algorithms,we propose the use of the vehicle tracking method based on neighborhood,to effectively eliminate error detection on the vehicle detection and to assure the stability and accuracy of the nighttime vehicle detection system in different scenarios.Finally,this algorithm is implemented in Visual Studio 2010 programming environment in Windows7 system and has achieved detection and tracking of vehicles in the nighttime scene.The experiment results on the vehicle videos with different brightness have proved that the proposed algorithm in this paper has a higher detection rate and better brightness adaptability,and also better real-time performance.