Gait Recognition Based on Key Point Motion Trajectory Modeling

doi:10.12068/j.issn.1005-3026.2024.01.005

Abstract

Abstract:

Gait information is a new biological characteristic with wide application prospects in medical and forensic fields， making it a hot spot in current research. Although researchers have proposed a variety of gait recognition methods， there are still some problems such as poor adaptability， overly complex feature description， and lack of interpretability. To solve this problem， firstly， the three-frame difference algorithm is improved to extract the human contour from video images. Then， a central structure model of human body is established based on the human body contour diagram， allowing for the identification of key points and the modeling of trajectory curves based on their locations in the video. Finally， a new gait feature description method is proposed using the previous curve model， with appropriate model parameters selected as gait feature vectors and suitable classification methods chosen for gait recognition and classification. Experimental results show that the proposed gait feature expression based on the trajectory model of key points can describe human gait information well and the recognition rate is relatively high.

Key words: gait recognition, contour extraction, human skeleton extraction, key point motion trajectory

CLC Number:

TP 20

Jiu-qiang XU, Xiao-xiao ZHAO, Long-fei QIAN. Gait Recognition Based on Key Point Motion Trajectory Modeling[J]. Journal of Northeastern University(Natural Science), 2024, 45(1): 33-39.

Figures/Tables 15

Fig. 1 Experimental results of three frame

Fig. 2 Length‐width ratio of the silhouette

Fig. 3 Eight neighborhood distribution of P0

Fig. 4 Refinement processing results

Fig. 5 Schematic diagram of human scale model

Fig. 6 Human body refined images and key point

Fig. 7 Examples of human body key points trajectory

Fig. 8 Fitting curves of step width，head and hip joint

Fig. 9 Gait cycle curves

Fig. 10 Human head movement trajectory

Fig. 11 Human step length change

Table 1 Recognition accuracy of different k values in the same state

k值	行走状态
k值	Nm	Cl	Bg
平均值	89.36	74.5	72.5
1	88.7	70	68
3	96.15	82	80
5	90.75	75	73
7	85.00	71	69

Table 2 Recognition accuracy of different k values

训练集	测试集	k=1	k=3	k=5	平均值
Nm	Cl	65	74	69	71
Nm	Bg	61	70	64	65

Table 3 Comparison of recognition accuracy of different algorithms

算法	正常行走	相同状态（Bg,Cl）	跨状态
文献［13］	88	—	45
文献［14］	95	80	—
本文	96.15	82	68

Table 4 Computational time of different algorithms

算法	计算时间/ms
GEI	87.0
文献［15］	3.2
本文	3.5

References 6

1	张引兵，刘楠楠，张力.身份认证技术综述［J］.电脑知识与技术，2011，7（9）：2014-2016.
	Zhang Yin‐bing， Liu Nan‐nan， Zhang Li.Authentication technology overview［J］.Computer Knowledge and Technology，2011，7（9）：2014-2016.
2	王蕴红，谭铁牛.现代身份鉴别新技术：生物特征识别技术［J］.中国基础科学，2000，2（9）：4-10.
	Wang Yun‐hong， Tan Tie‐niu.Modern identification technology：biometrics based personal identification［J］.China Basic Science，2000，2（9）：4-10.
3	Cunado D， Nixon M S， Carter J N.Using gait as a biometric，via phase‐weighted magnitude spectra［C］//Proceeding of International Conference on Audio‐and Video‐Based Biometric Person Authentication.Berlin，1997：95-102.
4	Zhao G Y， Liu G Y， Li H，et al.3D gait recognition using multiple cameras［C］//7th International Conference on Automatic Face and Gesture Recognition （FGR06）.Southampton，2006：529-534.
5	Han J， Bhanu B.Individual recognition using gait energy image［J］.IEEE Transactions on Pattern Analysis and Machine Intelligence，2006，28（2）：316-322.
6	Ben X Y， Gong C， Zhang P，et al.Coupled bilinear discriminant projection for cross‐view gait recognition［J］.IEEE Transactions on Circuits and Systems for Video Technology，2020，30（3）：734-747.