Preceding Vehicle Detection System Based On The Monocular Vision In Nighttime

Due to the critical traffic safety problem that the human society is confronted with in recent years, prevention-oriented active safety technology of modern vehicle has received more and more attention. The vehicle forward collision warning system (FCW) is the important part of Driving Assistance Systems (DAS), which has been believed to improve driving safety effectively. In order to achieve ideal performance of FCW, it is essential to obtain appropriate recognition of driving environment information. To solve the problems such as the recognition technology does not have good adaptability to the illumination environmental variation in nighttime environment, the FCW algorithm seldom considers the influence of the driver characteristics, etc, a FCW system based on the monocular vision in nighttime is constructed in this thesis, and the key technologies related to the system are studied.Firstly, in order to improve the vehicle detection in the nighttime environment, a brightness cumulative histogram based vehicle detection algorithm is proposed in this paper, which detects from the front vehicles via the highlight feature of the taillights. First, the initial threshold of Otsu method is obtained from a number of taillights statistical information. Then bright objects are extracted from images, based on the improved Otsu threshold method in brightness cumulative histogram. Finally, combining the characteristics such as the shape, position and color of taillights to select and pair them. Then the front vehicles can be detected by the paired taillights.Then, to enhance accuracy and promptness of vehicle detection method, we further describe a method to detected target vehicles are tracked using a time series analysis model, which is used to predict potential vehicle regions. Time series analysis model is initialized by the detection result in global scope of the image. In addition, we propose an improved detection method adapt to the tracking regions.In order to analysis the data of the driver car following characteristics and operation habits for the research on Forward Collision Warning System, the experimental data is dealt and analyzed efficiently. The FCW algorithm based on time-to-collision are designed according to the results of driver characteristics analysis. Then on the basis of detecting the forward vehicles, the distance and the relative speed between host vehicle and the target vehicle are calculated.Experiment is carried out on real experiment data captured by driver in road to verify the function and reliability of vehicle detection algorithm and FCW algorithm. Experimental results demonstrate the accuracy and robustness of the approach on vehicle detection at night, and demonstrate the FCW algorithm can accord with the driver car following characteristics and operation habits, and improve its adaptability for drivers on the certain scenarios.