Research Of Preceding Vehicles Detection And Tracking Based On Monocular Vision

With the rapid development of highway transport, while bringing convenience to people’s lives, but also causes highway traffic safety problems becoming increasingly prominent. As an important part of intelligent transportation system, safety driving assistance system can effectively prevent the occurrence of traffic accidents and improve traffic safety. Detection and tracking of preceding vehicles is the core link of safety driving assistance system, which provides an important guarantee for information extraction and vehicle behavior analysis. Based on the analysis and comparison of various algorithms at home and abroad, this paper research and put forward the corresponding vehicle detection and tracking algorithm based on monocular vision. The main contents are as follows:1) The research background of preceding vehicles detection and tracking is discussed in detail. Elaborating common methods for preceding vehicles detection and tracking based on monocular vision, and analyzing the advantages and disadvantages of these methods, to prepare for the follow-up study.2) Researching preceding vehicle detection algorithm based on AdaBoost. Selecting Haar-like features as the image feature, using Gentle AdaBoost algorithm and cascade algorithm trains samples offline to generate a cascade structure vehicle classifier; during the detection process, geometric amplification of detection window is applied to scan the under detected image, then using vehicle classifier to classify the detection window, by integrating the results of each detection windows and the final position of the vehicle is acquired. Experimental results show that this method can effectively detect preceding vehicles, with a certain robustness, but meet real-time requirements generally.3) A preceding vehicle tracking algorithm based on improved TLD is proposed. Tracking-Learning-Detection(TLD) algorithm is a novel target tracking algorithm. It can learn object features quickly and track effectively with extremely limited prior knowledge given. The priori knowledge of preceding vehicles can be provided by vehicle detection algorithm, therefore, TLD algorithm is fully applicable to the preceding vehicle tracking. However, the tracker in TLD tracking algorithm selects the target feature points uniformly each time, which cannot ensure that each of the feature points can be reliably tracked. To solve this problem, an improved TLD algorithm is proposed, based on key feature point detection. The proposed algorithm can ensure that the selected feature points can be tracked reliably and correctly. Drift tracking is prevented and tracking accuracy is improved. Furthermore, online location prediction based on trajectory continuity is introduced into the TLD detector, which can not only ensure proper tracking but also reduce the detection range of the detector. What’s more, the computing speed is improved. Finally, the improved TLD algorithm is applied to track preceding vehicles. Experimental results show that the algorithm can effectively track preceding vehicles, and has good tracking performance in various cases which are difficult to deal with.