基于特征融合的目标动态身份识别方法

Target Dynamic Identity Recognition Method Based on Feature Fusion

人脸特征判别受摄像头安装位置与角度、目标活动以及光线变化等因素影响,识别准确度急剧下降,因此提出了一种基于人脸和行人特征融合的目标动态身份识别方法。使用特征融合方法,将人脸特征和行人特征首尾拼接得到一个融合特征,从而获得丰富的目标身份信息;使用匹配决策机制,通过求和对多个融合特征的匹配结果取最大值,输出概率最高的目标身份,从而降低误判单个特征目标身份的概率。在真实场景的监控图像中,该设计方法的识别准确度平均为87.7%,较人脸识别提高了40.0%。实验表明,人脸特征判别性不足时,补充行人特征可有效增强目标身份信息,且多个融合特征匹配结果综合决策可进一步提高识别准确度。

Affected by factors such as camera installation position and angle,target activity,and light changes,facial features are not sufficiently discriminative,leading to a sharp drop in recognition accuracy.In response to this challenge,a target dynamic identity recognition method based on face-pedestrian feature fusion is proposed.First of all,a feature fusion method is proposed,which combines the end-to-end facial features and pedestrian features to obtain a fusion feature to obtain richer target identity information.Secondly,a matching decision mechanism is proposed,the maximum matching result is obtained by summing the matching results of multiple fusion features,and the target identity with the highest probability is output,which reduces the probability of misidentification of the target identity of a single feature.In real-world monitoring scenario images,the average recognition accuracy of the design method is 87.7%,which is 40.0%higher than that of face recognition.Experiments show that the combination of pedestrian features can effectively enhance the target identity information when the facial features are not discriminative enough,and the integrated decision of multiple fusion feature matching results can further improve the recognition accuracy.