﻿ 复杂场景下的运动目标识别算法
 计算机系统应用  2018, Vol. 27 Issue (8): 193-197 PDF

Recognition Algorithm of Moving Target in Complicated Scenes
GONG Fa-Ming, LI Xiao-Ran, MA Yu-Hui
College of Computer & Communication Engineering, China University of Petroleum, Qingdao 266580, China
Foundation item: Special Project for Innovative Work Methods of Ministry of Science and Technology (2015IM01030)
Abstract: Target recognition is the basic purpose of computer vision, it is also one of the key components in the field of artificial intelligence.With the advent of the information age, the popularity of video capture tools, massive video data to human identification has brought great challenges. At this stage, video recognition technology has been widely used in simple scenes such as intelligent transportation field and production quality inspection field. How to realize the target recognition and detection from complex scenes has become a more important and difficult issue. In response to this problem, this paper presents a moving target recognition algorithm in complex scenes. First, an improved optical flow algorithm is proposed to mark the moving target region quickly by time series and spatial pixel changes; Secondly, the sliding window of the target area is detected to match the model of each part of the human body, and the feedback information is modeled by a tree structure. Through experiments, this method can detect faster than detection algorithm based on depth learning while ensuring high accuracy, and can meet the requirements of real-time monitoring.
Key words: computer vision     optical flow algorithm     sliding detection     model matching     target recognition

1 引言

 图 1 目标识别算法流程图

2 复杂场景下的运动目标检测算法 2.1 预处理和光流检测

 $I(x,y,t) = I(s + u\delta t,y + v\delta t,t + \delta t)$ (1)

 ${E_D}(u,v) = \sum\limits_{(x,y) \in \Re } {{{({I_x}u + {I_y}v + {I_t})}^2}}$ (2)

 ${E_S}(u,v) = u_x^2 + u_y^2 + v_x^2 + v_y^2$ (3)

 $E(u) = {E_D}(u) + \lambda {E_{{r}}}(u)$ (4)
 图 2 改进的光流算法流程图

2.2 滑动窗口模型匹配法

 $S({p_i}) = \sum\limits_{i \in V} {\varpi _i^{{t_i}} \cdot \phi (I,{p_i})}$ (5)

 $S(t) = S(t) + \sum\limits_{i \in V} {\varpi _i^{{t_i}} \cdot \phi (I,{p_i})} + \sum\limits_{i,j \in E} {\varpi _{ij}^{{t_i},{t_j}} \cdot \psi ({p_i} - {p_j})}$ (6)

 ${{p = [}}{{{p}}_1}{{,}}{{{p}}_2}{\rm{,}} \cdots{\rm{,}}{{{p}}_N}{\rm{]}} \in {R^{D \times N}}$ (7)

 ${c_i} = \mathop {\arg \min }\limits_{c \in {R^M}} \left\| {{p_i} - {F_c}} \right\| + \lambda {\left\| {{d_i}\Delta c} \right\|^2},\;\;{\rm{s}}{\rm{.t}}{\rm{.}}\;\;{{\rm{1}}^{\rm{T}}}c = 1$ (8)

 ${d_i} = \exp ( - \frac{{{{((D({p_i},F) - {D_M})/{D_M})}^2}}}{{2{\sigma ^2}}})$ (9)

 ${D_m} = \min \{ D({p_i},{f_j}),D({p_i},{f_2}), \cdot \cdot \cdot ,D({p_i},{f_M})\}$ (10)

2.3 目标识别

3 实验结果与分析

 图 3 复杂场景下的目标检测法所得结果图

 图 4 卷积神经网络法所得结果图

(1) 在相同的数据检测中, 本文提出的方法在检测效率上要优于基于深度学习的检测算法.

(2) 在相同的数据检测中, 本文提出的方法在检测质量上略低于基于深度学习的检测算法, 但在资源的消耗程度上要优于深度学习检测算法.

4 结论和未来工作

 [1] Mishne G, Talmon R, Cohen I. Graph-based supervised automatic target detection. IEEE Transactions on Geoscience and Remote Sensing, 2015, 53(5): 2738-2754. DOI:10.1109/TGRS.2014.2364333 [2] 肖军, 朱世鹏, 黄杭, 等. 基于光流法的运动目标检测与跟踪算法. 东北大学学报(自然科学版), 2016, 37(6): 770-774. DOI:10.3969/j.issn.1005-3026.2016.06.003 [3] 屈晶晶, 辛云宏. 连续帧间差分与背景差分相融合的运动目标检测方法. 光子学报, 2014, 43(7): 213-220. [4] Dalal N, Triggs B. Histograms of oriented gradients for human detection. IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Diego, CA, USA. 2005. 886–893. [5] Yan JJ, Lei Z, Wen LY, et al. The fastest deformable part model for object detection. IEEE Conference on Computer Vision and Pattern Recognition. Columbus, OH, USA. 2014. 2497–2504. [6] Ramakrishnan SK, Pal A, Sharma G, et al. An empirical evaluation of visual question answering for novel objects. IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, HI, USA. 2017. 7312–7321. [7] 邵奇可, 李路, 周宇, 等. 一种基于滑动窗口优化算法的行人检测算法. 浙江工业大学学报, 2015, 43(2): 212-216. DOI:10.3969/j.issn.1006-4303.2015.02.019 [8] 杨涛, 李静, 潘泉, 等. 基于场景模型与统计学习的鲁棒行人检测算法. 自动化学报, 2010, 36(4): 499-508. [9] Belagiannis V, Amin S, Andriluka M, et al. 3D pictorial structures for multiple human pose estimation. Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition. Columbus, OH, USA. 2014. 1669–1676. [10] Tang SY, Andriluka M, Schiele B. Detection and tracking of occluded people. International Journal of Computer Vision, 2014, 110(1): 58-69. DOI:10.1007/s11263-013-0664-6 [11] Wen ZW, Yin WT, Zhang Y. Solving a low-rank factorization model for matrix completion by a nonlinear successive over-relaxation algorithm. Mathematical Programming Computation, 2012, 4(4): 333-361. DOI:10.1007/s12532-012-0044-1 [12] Wei SE, Ramakrishna V, Kanade T, et al. Convolutional pose machines. arXiv: 1602.00134, 2016: 4724–4732. [13] 方智文, 曹治国, 朱磊. 基于梯度角度的直方图局部特征描述子的图像匹配算法. 计算机应用, 2015, 35(4): 1079-1083. DOI:10.11772/j.issn.1001-9081.2015.04.1079 [14] Rother C, Minka T, Blake A, et al. Cosegmentation of image pairs by histogram matching - Incorporating a global constraint into MRFs. Proceedings of 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. New York, NY, USA. 2006. 993–1000. [15] 刘丽, 匡纲要. 图像纹理特征提取方法综述. 中国图象图形学报, 2009, 14(4): 622-635. DOI:10.11834/jig.20090409