School of Transportation Engineering, Tongji University, Shanghai 201804, China;College of Computer Science and Engineering, Jiangsu University of Science and Technology, Zhenjiang 212003, China 在期刊界中查找 在百度中查找 在本站中查找
In order to detect the traffic incidents occurred on highway and reduce the loss brought by traffic incident, this paper presents an improved BP_AdaBoost algorithm based on genetic algorithm for traffic incident detection. The inputs of BP(Back Propagation)neural network value are vehicle quantity, velocity and occupancy in upstream and downstream of highway. Genetic algorithm is used for each BP neural network classification model for optimizing weights and thresholds due to its performance of global searching. The optimized BP neural network model is applied as a new weak classifier,then through the AdaBoost algorithm,many of these new weak classifier is composed as strong classifier model.This improved algorithm is validated with real data from Tokyo expressway ultra sonic sensors. The experimental results show that the algorithm can improve the performance of BP weak classifier. The detection rate of improved BP_AdaBoost algorithm is up to 97%, and false alarm rate is lower to 3.34%. Experiment indicate that the algorithm is suitable for detecting highway traffic incidents.