CAJ | 학술논문

在非结构环境下，基于视频检测目标车速为目的的关键问题是：如何利用视频序列准确、高效地得到运动目标的实时信息。基于视觉感知机制的车速检测是一种新方法，该方法以视觉机制为启发，首先从图像中学习初级视皮层细胞超完备基；然后基于超完备基视觉计算模型，以神经元响应系数表征目标，并根据背景与目标响应系数差值与动态阈值的比较识别运动目标；通过迭代实现目标跟踪；在目标跟踪的基础上，根据不同视频序列中车辆特征匹配法得到车辆的位移，最后实现车速检测。实验结果表明，该算法能够在实时条件下检测实际道路场景的不同类型车辆速度，达到96．55％以上的准确率，并且与传统方法相比较提高了车速检测的准确性和鲁棒性。
재비결구배경하，기우시빈검측목표차속위목적적관건문제시：여하이용시빈서렬준학、고효지득도운동목표적실시신식。기우시각감지궤제적차속검측시일충신방법，해방법이시각궤제위계발，수선종도상중학습초급시피층세포초완비기；연후기우초완비기시각계산모형，이신경원향응계수표정목표，병근거배경여목표향응계수차치여동태역치적비교식별운동목표；통과질대실현목표근종；재목표근종적기출상，근거불동시빈서렬중차량특정필배법득도차량적위이，최후실현차속검측。실험결과표명，해산법능구재실시조건하검측실제도로장경적불동류형차량속도，체도96．55％이상적준학솔，병차여전통방법상비교제고료차속검측적준학성화로봉성。
The key issue of the purpose of detecting target vehicle speed based on video is how to obtain the accurate and effi-cient real-time information of moving targets using the video sequence in unstructured environment.Vehicle speed detection based on visual perception mechanism is a kind of new method.Visual mechanism inspired the method,first of all it should learn the over-complete characterization of primary visual cortex cells from nature images.Then it compared with the dynamic threshold to identify the moving target based on over-complete visual calculation model according to neuron response coefficient characterization and the difference of neuronal response coefficient between background images and goal images.Next,it a-chieved target tracking through the iterative method.Compared with the traditional method it got the displacement of the vehicle according to the matching method of characteristics in different video sequences on the basis of target tracking.Finally it could realize speed detection by this algorithm.Experimental results show that this algorithm can test the vehicle speed of different types in the actual road scene under real-time conditions and reach more than 96.55% accuracy.The algorithm improves the vehicle speed detection in accuracy and robustness compared with the traditional method.