| With the rapid development of road traffic, wide attention has been paid to traffic safety, and the researches of vehicle active safety system have became increasingly active in the last decade. Now advanced Driver Assitance System based on vision becomes an important research direction of vehicle active safety technologies, since the visual sensors are in low price and high fit with drivers. Early warning system designed to assist the driver assitance system is researched in this thesis, including the car detection and tracking, lane line detection and anti-collision warning algorithm. The main contents of this Thesis can be divided into four parts.1. This thesis studies a feature extraction method based on compressed sensing which can reduce the characteristic dimension effectively and retain almost all the vehicle information. Then, a soft cascade classifier based on linear SVM is trained with the help of Bootstrap. After that, image pyramid fast calculation algorithm and window search strategy which based on particle filter are introduced to real-time vehicle detection.2. This thesis realizes an algorithm for lane line detection based on learning. The road area is obtained by the vanishing point and horizon position, and the detection area is restricted on the road area. With the fact that lane line cannot change suddenly, a strategy aims at getting more stable detection result is introduced.3. Combining with the kalman filter and particle filter, a new tracking algorithm is realized to improve the tracking efficiency. Using the output of particle filter as the input of kalman filter makes full advantage of both image characteristics and movement trend.4. With the help of the research on the acceleration variation, an anti-collision warning strategy is designed based on the relative velocity. Combined with the distance measurement algorithm, a vehicle anti-collision warning system is designed. In the end, the design of anti-collision warning demo is completed. |
PDF Full Text Request