摘要
在辅助驾驶中对行人检测的实时性要求较高,然而现有方法的计算复杂度普遍较高,难以满足实时应用的需求。该文提出了一种基于Ada Boost算法的行人检测与预警方法。行人探测器由Opencv训练,并且训练12级行人探测器。通过在处理器上使用检测器,实现视频图像的行人检测功能。另外,通过使用可变步长和分区扫描跟踪方法,有效提高了检测率,实时检测的要求基本达到。采用均匀化预处理和多尺度融合技术提高了检测过程中检测率;通过直方图均衡办法有效地剔除了光影等因素对测试结果的影响;在测试完成后,通过多尺度融合技术消除冗余部分的测试结果,最终不仅减低了误报率而且大大提高检测效率。
In the auxiliary driving pedestrian detection of real-time requirements are higher.However, the computational complexityof existing methods is generally high, which makes it difficult to meet the requirements of real-time applications.This paper presents a method based on Ada Boost algorithm for pedestrian detection and early warning.Pedestrian detectors are trained by Opencvand train 12-level pedestrian detectors. By using the detector on the processor, video image pedestrian detection. In addition,through the use of variable step and partition scanning and tracking methods, effectively improve the detection rate, real-time detection of the basic requirements.Homogenization pretreatment and multi-scale fusion technology to improve the detection rate inthe detection process;Histogram equalization method effectively eliminates the light and other factors on the test results;After thetest is completed, the redundant part of the test result is eliminated through the multi-scale fusion technology, and finally the detection rate is increased and the false alarm rate is reduced.
出处
《电脑知识与技术》
2018年第2X期174-177,共4页
Computer Knowledge and Technology
基金
河北省科技厅自筹项目(16210805)
河北省科技厅项目(16273705D)
河北省科技厅项目(16222101D-1)
