基于车载计算机的红外图像移动目标检测

Infrared Image Moving Object Detection Technology Based on Onboard Computer

环境感知系统作为无人车平台的重要组成部分,是路径规划与决策控制等功能实现的基础与前提。结合车载计算机的硬件基础,设计一种复杂场景下的红外图像移动目标检测系统。该系统采用客户端/服务端(B/S)架构,在客户端浏览器进行服务请求后,服务端使用改进的YOLOv2算法对车载Rapid输入输出(RapidIO)高速总线传输来的红外图像进行目标检测;通过维度重聚类和改进激活函数等方法,有效降低了漏检和误检率;为弥补红外图像对比度低、纹理特征弱的缺点,提出一种基于直方图均衡化的双边滤波图像增强算法,该预处理能够有效保持图像轮廓细节信息。结果表明:改进YOLOv2算法的实时检测准确度相对于原始算法提升了4.1%;图像预处理算法显著提高了检测准确率;该系统可为开拓无人车载红外图像的应用领域提供可靠的技术支撑。

As an important part of driverless vehicle platform,the environmental perception system is the basis and premise to realize as the path planning and decision control function.Based on the onboard computer,an infrared image moving object detection system in complex scenes is designed.The browser/server architecture is used in the detection system.After the client browser requests for services,the server applies the improved YOLOv2 algorithm to detect the infrared images transmitted by the onboard RapidIO high-speed bus in real time.the missed detection and error detection rates are effectively reduced by dimensional re-clustering and improved activation function.In order to make up for disadvantages of low contrast and weak texture features of infrared images,bilateral filtering image enhancement algorithm based on histogram equalization is proposed,which can effectively maintain the image contour details.The results show that the real-time detection accuracy of the improved YOLOv2 is increased by 4.1% compared with the original algrithm and the image preprocessing significantly improves the detection accuracy.