摘要
由于每个人走路方式不同,使得各人鞋底的磨损程度也不同,因此在很多案件中,法医都会通过"鞋印"来对嫌疑人进行侧写和识别.截至目前,多项研究已经证实,在人员识别方面,"鞋型"和"步态特征"确实是一类非常有用的工具.不过,单一特征识别设备存在明显缺点,如操作复杂以及抗干扰性难以达标等.为解决上述"短板",文章提出了一种基于BP神经网络联合"鞋型"和"足底压力特征"的身份识别系统.通过比对3 种不同的足底压力数据积累方法,研究发现,所提系统在无"噪声"的情况下能够达到最佳精确度(89%),而当"添加"10%的噪声像素时,其精确度亦能够达到 74%.所得结果证实,该系统在人员识别方面的性能确实令人满意.
Abstract
A shoe print is a unique feature of each shoe,and is used in the forensic case to identify the suspect.Many kinds of research shows it is effective in identifying people with shoe shape and gait features.To solve the drawbacks of complex equipment and poor immunity to interference caused by the use of one feature alone,this paper studied a people identification system based on BP neural network and combined both shape and plantar pressure features.By comparing the 3 different pressure data accumulation methods,the system achieved a best accuracy of 89%without noise and 74%with about 10%noise pixels added.The result of the experiment demonstrated the effectiveness of this system.
NETL
NSTL
万方数据