Research On Vehicle Target Risk Identification And Risk Assessment Based On Machine Vision

Currently, road traffic safety problem has been getting more attention, driver assistance system is an important vehicle active safety technology, and it is also the key research direction of the intelligent transportation and smart vehicles. The dangerous object which is front of the vehicle, its’ recognition is an important part of driver assistance systems, using the object detection and tracking technology to identify the object in front of the vehicles, and determining its’level of risk, it can greatly reduce the degree of traffic accidents, and improve vehicle safety. In a complex road traffic environment, the video background is dynamic, which leads to many problems when using the current detection and tracking algorithms, researching on these issues, and estimating the tracked object’s motion parameters, assessing their risks and warning.In terms of object detection, this paper uses the pedestrian and vehicle detection algorithm based on the characteristics of HOG. According to the samples’HOG (Histogram of Oriented Gradient) features, it using the SVM classifier to learn and classification. In algorithm improvements, this paper uses principal component analysis to reduce the dimension of HOG features, and for the video image, using the region segmentation, determining the objects’possible existence region, then, using the sliding window to detect possible area, the identification algorithm can basically meet the requirements of real-time online processing. The accuracy of detection algorithm is also improved, using gray symmetry measure and local entropy calculation can exclude the corresponding false detection object. In object tracking, this paper, using the object tracking algorithm based on particle filter, and according to the steps of the particle filter algorithm to achieve target tracking, and making a brief analysis for the tracking results.This paper estimating the tracked object’speed, motion direction and distance in the premise of target detection and tracking. The object’s direction estimation using the method based on historical images, determining the motion module’s overall gradient direction, then, achieving the object’s overall.motion direction. The object’s speed estimation using conversion ratio which is obtained from the object’s image velocity and actual speed, the object’s image velocity is obtained by target centroid displacement and frame rate. The object’s distance estimation using camera’ pinhole imaging principle to calculate, firstly, calibrating the internal parameters of the camera, according to the calibration results, combining with the known parameter, calculating the object relative distance. Finally, using the fuzzy comprehensive evaluation to assess the target’s collected motion parameters, determining the object’s risk level, and displaying on the video image.Finally, using OpenCV2.4.2and MS VC++2008to program the road target detection, tracking and motion analysis software, the software can accurately detect pedestrians and vehicles objects, and estimate the object motion parameter, preliminarily judge its risks.This article studies the dangerous objects’ recognition and risk assessment which are front of the vehicle, the subject’ result has a certain significance to the study of the vehicle active safety technology.