夜间行人图像的智能识别方法改进

Improvement of Intelligent Identification Method for the Pedestrian Images at Night

在夜视条件下对行人识别、检测过程中会产生拍摄视角变化的复杂性、行人衣着和姿态的多样性、成像尺度的实时性不足等问题,尤其是对人脸、指纹等细节特征的识别尤为困难。传统方法在夜间对行人的识别缺乏实效性和精准性,提出基于优化的IE_HOG特征方法,在夜间能更精确地描述人体轮廓特征,结合Radon变换动态特征提取技术,对行人在夜间的步态图像进行智能识别。试验证明提出的方法能够在夜间更好地对行人轮廓特征和步态变换情况进行精准的识别,同时降低目标敏感性,排除树木、建筑物等遮档物的干扰,提高鲁棒性,图像识别率可以达到94.39%。

It can lead to complexity of view variation, diversity of pedestrian clothing and posture and poor instantaneity of imaging scale to identify pedestrian under the night vision condition in the progress of detection. The detail feature identification of face and fingerprint is especially difficult. Traditional method is short of effectiveness and precision of pedestrian identification at night. In this paper, we propose a modified IE_HOG feature method. It described body contour feature more precisely. We integrate the Radon transformation dynamic feature extraction technique to identify gait image of pedestrian at night intelligently. The experiment results show that the method mentioned above can identify the contour feature and gait variation state precisely. It can reduce the target sensibility, eliminate the interference of trees and construction and improve the robustness. The rate of image recognition can reach 94. 39%.