多证据信息融合的行人运动状态识别模型

Motion State Recognition Model of Pedestrian Under Multi-Evidence Information Fusion

采用粒子滤波的方式对人的运动情况进行分析的时候,进行数据采集的时候经常会产生数据不够,造成系统运行结果差强人意,鲁棒特性较差。基于这样的问题,算法结合了色彩、空间和细节特征等多方面的数据信息来完成外观模型的建立,在对其时域特性和详细特征进行分析之后,使用RGB颜色模式完成对应的模型;经过LBP处理之后,对进行优化的TD-LBP模型进行进一步说明,除去亮度小范围变化给纹理区分带来的影响;并且,引入了空间概念,目的在于依据人的特点,同时通过亮度的方式将一定的区域进行进一步划分,成为3个子区域,接着把不同子区域内的证据信息进行提取。经过实验证明之后,采用的这个模型在综合了直方图作为辅助之后,能够完成复杂情况下的行人跟踪分析,所采用的TD-LBP算法也比LBP算法更加的精准。

When the particle filter algorithm is applied to the pedestrian motion state estimation,it often results in low efficiency and poor robustness because of the loss of produced data in the sampling phase.In order to solve the above problems,the algorithm combining with various aspects of information in color,space and detail feature to complete the construction of pedestrian appearance model.Fully analyzing the real-time and specific characteristics,it uses RGB color space to construct the corresponding model.Then further clarify the optimized TD-LBP algorithm model,and eliminate the texture distinguish interference produced by the slight fluctuations in pixel brightness.In addition,the algorithm also introduces the space concept in order to divide the pedestrian area into three sub-areas by the brightness information with the pedestrian characteristics.Then extract the evidence information of the different sub-areas.After experimental verification,the multi-evidence information fusion model proposed by this paper can effectively solve the problem of pedestrian tracking in complex conditions under the conjunction with the integral histogram algorithm,at the same time,the TD-LBP algorithms involved in it is more accurate in recognition rate than the general traditional LBP algorithm.