摘要：Obstacle detection is the key technology of environment perception for intelligent vehicle. To guaranteesafe operation of an unmanned vehicle the fusion method of spatial information from lidar and machine vision is studied. The projection of the bounding box generated by lidar on machine vision image is designed. The detection area of obstacleis constructed. Moreover, the method of strong-classifiercascade connection is used to build a classifier. Specially, theHaar-like and HOG features within a huge amount of data are characterized based on AdaBoost algorithm. To achieveclassification and recognition of the obstacles, the obstacle detection areas generated on the image are fused with the designed cascade classifier, and the effectiveness of theproposed fusion method is validated. Test results show that the obstacle detection method based on fusion of laser radar and machine vision shows higher detection accuracy. Under good weather conditions, compared with the detection method basedon laser radar alone and based on machine vision alone, theproposed method increases the detection rate of vehicle obstacle by 18.33% and 12.74%, and increase the detection rate of pedestrian by 17.92% and 12.56%, respectively. Under the rainy weather, the detection rate of vehicle obstacle is improvedby 38.44% and 14.28%, and the detection rate of pedestrian is enhanced by 29.34% and 15.84%, respectively.
关键词：unmanned vehicle; obstacle detection; data fusion; laser radar; machine vision
