王超, 纪明, 王娇颖, 姜文涛, 解静, 李涛, 高雅. 一种基于数字图像的多目标检测方法[J]. 应用光学, 2015, 36(3): 386-390. DOI: 10.5768/JAO201536.0302001
引用本文: 王超, 纪明, 王娇颖, 姜文涛, 解静, 李涛, 高雅. 一种基于数字图像的多目标检测方法[J]. 应用光学, 2015, 36(3): 386-390. DOI: 10.5768/JAO201536.0302001
Wang Chao, Ji Ming, Wang Jiao-ying, Jiang Wen-tao, Xie Jing, Li Tao, Gao Ya. Multi-target detecting algorithm based on digital image[J]. Journal of Applied Optics, 2015, 36(3): 386-390. DOI: 10.5768/JAO201536.0302001
Citation: Wang Chao, Ji Ming, Wang Jiao-ying, Jiang Wen-tao, Xie Jing, Li Tao, Gao Ya. Multi-target detecting algorithm based on digital image[J]. Journal of Applied Optics, 2015, 36(3): 386-390. DOI: 10.5768/JAO201536.0302001

一种基于数字图像的多目标检测方法

Multi-target detecting algorithm based on digital image

  • 摘要: 为使武器系统具备同时对多个目标进行精确打击的能力,在光电系统中可采用多目标视频跟踪器辅助激光照射器、伺服稳定平台实现多个潜在目标同时捕获跟踪并打击。研究了一种运行于多目标视频跟踪器的目标检测方法,针对数字图像分辨率高、数据量大及难以在嵌入式系统中实时运行的难点,基于TMS320C6455 DSP处理器,提出基于小波金字塔的全局运动光流估计算法图像配准实现运动图像的背景补偿以获取差分图像,相比传统的块匹配、灰度投影配准及基于特征点的配准算法,具有配准精度高与可嵌入式系统实时处理等优点,在差分图像中采用区域生长结合管道滤波算法提取图像中多个运动目标。经实验验证,该方法在复杂地面场景对汽车、自行车及行人目标检出率可达95%,计算时间仅为25 ms,具有良好的实时性和检测效果。

     

    Abstract: In order to improve the capacity of weapons system to attack multiple targets precisely at the same time, the multi-target tracker can be used to assist laser and servo platform to capture and track multiple potential targets in electro-optical system. A novel multi-target detecting algorithm run in multitarget tracker was studied. According to the difficulty of processing timely in embedded system for the high resolution of digital image and mountainous data flow, based on TMS320C6455 digital signal processing (DSP), a global optical flow estimate image mosaic algorithm based on wavelet pyramid decomposition was proposed to achieve background compensation in moving image to get difference image.Compared to traditional block matching, gray-scale projection matching and key-points relevant matching algorithm, the new method is more accurate and timely running. Region growth combined with tunnel filter algorithm was carried out to acquire multiple final targets in difference image. The experiment verified that the method can achieve 95% detection rate and cost only 25 ms for automobile, bicycle and pedestrian in complex ground, which has good real-time performance and detecting effect.

     

/

返回文章
返回