基于序列图像的实时人流检测与识别算法研究

针对可见光下人流检测、识别算法中存在的运动目标分割准确率低、识别效果差等问题,提出一种新的跟踪与识别方法.首先利用序列图像中运动目标时空一致性,将帧间二阶差分(SODP)与边缘检测相结合进行运动目标分割;再根据行人运动模型和运动目标局部性特征,通过粗采样方法快速提取跟踪特征向量;利用运动目标轮廓投影比、形状因子等特征分量,并构造基于人工神经网络的运动目标分类器进行识别.通过对大型商场进行的实际测试表明:该方法在运行效率、识别准确率方面均取得满意结果.同时,算法对于光线、阴影和人流变化等外界因素的影响,具有较强适应性.

Real Time Detection and Recognition of Passenger Flow Based on Image Sequences

A new pedestrian tracking and recognizing method is presented aiming at solving the problems in tracking and accounting pedestrian flow in visible light, such as low accuracy in dividing moving object and poor recognizing effect. Firstly, the moving object is divided by combining the SODP with edge detection and measure according to the consistency of the moving objects in the visual flow. Secondly, the tracking feature vectors are extracted rapidly with rough sampling based on the pedestrian moving model and the part feature of moving object. And the satisfying accuracy and recognizing speed are obtained when a pattern recognizes moving objects with features such as projection ratio of moving object outline and figure factors, and with the moving object assorting tool based on artificial neural network. The method is used in real time tracking and accounting the pedestrian flow in large malls, and the practical test results indicate that satisfying effect is obtained in both processing efficiency and recognizing accuracy with this method. Moreover, it has good adaptability to external influence such as light of the test spot, shadow of pedestrian, and change of pedestrian flow.