Research On Driving Environment Detection Applied To Driving Assistance

The important research contents of assistant driving technology are traffic sign detection and pedestrian detection.However,the real driving environment is very complex.The factors,such as the illumination intensity,background complexity,pedestrian posture difference and so on,may make the detection barely satisfactory.Traditional traffic sign detection algorithm based on color is sensitive to illumination,which leads to barrier of selecting color threshold,low recognition rate and poor robustness.Consequently,a traffic sign detection algorithm based on visual attention mechanism and geometric shape is proposed.The Region of Interest(ROI)of image is located by using the improved visual attention model,followed by the pretreatment of shape detection that removes the noise interference.And the fine detection of the traffic sign is realized according to its geometrical characteristics.In addition,the pixel duty ratio and improved HU invariant moments of the traffic signs' internal images are extracted.Simultaneously,the BP neural network structure which is applicable for the recognition requirements is constructed to realize the accurate identification of the traffic signs.A pedestrian detection algorithm based on the fusion feature and cascade classifier is proposed in order to solve the problem that the pedestrian detection rate is not high.Four rectangular features are added to improve the contour features based on the Haar-like features.The gradient amplitude and direction of the pixel points in the diagonal direction are calculated on the basis of the HOG feature extraction algorithm,and the HOG feature is accelerated by the integral graph.The two features are combined to construct the Adaboost cascade classifier,which assists with the pedestrian positive and negative samples to train the classifier to achieve the best pedestrian detection effect.The results detected in real driving environment mirror that the traffic sign detection and pedestrian detection have higher accuracy.In conclusion,the data demonstrate that the algorithm still has strong robustness and good detection performance in the condition of the interference factors,illumination changes and complex background.