Target Detection And Distance Measurement Based On Monocular Vision

With the rapid increase of car parc,traffic environment has become more and more complex.In order to improve the active safety and reliability of vehicles,reduce the probability of traffic accident,the research and development of advance collision warning system has important social significance.Monocular vision was introduced in this paper,vadvance anti-collision warning system is researched under the driving state.The anti-forward collision warning system advance anti-collision warning system detects the front vehicle firstly,and then measures the distance between the two vhicles.It can avoid traffic accients when the driver's attention drops.The vehicle in the image is detected by the target detection algorithm based on AdaBoost classification,and the principle and process of the algorithm are in the second chapter,the vehicle in the image is detected by the target detection algorithm based on AdaBoost classification,and the principle and process of the algorithm are introduced.Since the vehicle tracking algorithm which takes less time tracks target in the target vehicle's image neighborhood range,the real-time performance of target detection is increased by combining vehicle detection and vehicle tracking in the follow-up section.This thesis proposes improved Gauss hybrid model based on target vehicle on the context of fluctuation in image background.Dynamically updating with the real-time changes of the target vehicle pixels,this model can distinguish target vehicle and dynamic background.What's more,this model can modify the image area and position of the target vehicle.Experiments show that the Gauss mixture model of target vehicle has certain discrimination ability to distinguish vehicle and background.Considering the easy appearance of tracking drift in Meanshift tracking algorithm,this thesis proposes a modified algorithm by combining Gauss hybrid model with Meanshift tracking algorithm.With full use of vehicle color information,the algorithm can well shield the interference of background information in the iterative process of Meanshift tracking algorithm.Synthetic and real experiments demonstrate high accuracy and stability of the improved algorithm.Aimed at the situation of losing target that the target vehicle suddenly speeding out of tracking area or being occluded,the particle filter tracking algorithm is introduced to achieve the target vehicle repositioning.Experiments show that the combination of particle filter and tracking algorithm can well deal with the temporary lost of target vehicles.In the aspect of monocular measurement,measurement algorithm based on geometric relations is used,and the specific experiments is carried out.Experimental results show that the measurement range of up to 45m.Finally,the research work carried out is summarized,and the future work is prospected.