半自主在线学习目标检测系统

摘要

针对不同监控场景，不同成像条件下目标姿态变化较大的问题，提出一种具有半自主学习能力的目标检测系统。该系统能在执行检测任务的同时，通过快速的半自主学习提高检测性能。系统包括了目标检测模块及在线学习模块。为满足系统在线学习需求，提出随机蕨分类器的在线学习方法，使目标检测模块可持续自我更新，提高检测性能。通过半自主在线学习框架使整个学习过程不需准备初始训练样本集，只需框选一个待检测目标即可进行自适应学习，逐渐提高检测性能。实验表明，该方法在多种监控场景中均有较强的自适应能力和较好的目标检测效果。%Since the object attitude has great variation in different monitoring scenes and different imaging conditions,an object detection system with semi⁃autonomous learning ability is proposed. The system can improve the detection performance by means of fast semi⁃autonomous learning while executing the detection task. The system is composed of object detection module and online learning module. To satisfy the requirement of system online learning,the online learning method of random fern classifier is proposed. It can sustainably self⁃renewal the object detection module,and improve the detection performance. The whole learning process by needn′t prepare the initial training samples semi⁃autonomous learning framework,and only select a detected object to perform the adaptive learning,so the detection performance is improved gradually. The experimental results show that the method has strong adaptive capability and high detection rate.