特征点辅助的时空上下文目标跟踪与定位

发布时间：2018-02-26 12:37

本文关键词： 目标跟踪时空上下文特征点跟踪双目视觉目标测量　出处：《仪器仪表学报》2017年11期 　论文类型：期刊论文

【摘要】：针对动态目标跟踪中快速运动和目标遮挡而跟踪失败问题,提出了一种特征点辅助的时空上下文跟踪算法。首先提取目标特征点,通过特征点匹配和光流跟踪方法进行目标追踪,获得目标预估位置;其次,建立特征点变化率和时空上下文模型更新率关系模型,实时调控更新率,防止引入错误信息;最后,在预估位置区域内,构建局部上下文外观模型,计算与时空上下文模型的相关性获取置信图,进一步精确定位目标。算法在一组测试视频集中进行验证,相比目前4种主流算法(平均跟踪成功率最高为60%,平均跟踪误差最小为26.14 pixel),本算法综合性能达到最优,平均跟踪成功率为90%,平均跟踪误差为7.47 pixel,平均跟踪速率25.31 f/s。在双目视觉移动机器人平台上对随机运动目标进行跟踪实验,在背景干扰、遮挡、目标旋转和快速运动等组合情况下,跟踪成功率97.4%,跟踪距离平均相对误差为4.05%。
[Abstract]:In order to solve the problem of fast motion and target occlusion in dynamic target tracking, a spatio-temporal context tracking algorithm assisted by feature points is proposed. Firstly, the target feature points are extracted. The target prediction position is obtained by feature point matching and optical flow tracking. Secondly, the relationship model of feature point change rate and spatio-temporal context model update rate is established to control the update rate in real time to prevent the introduction of error information. In the predicted location region, the local context appearance model is constructed, the correlation between the local context model and the spatio-temporal context model is calculated to obtain the confidence chart, and the target is further accurately located. The algorithm is verified in a set of test video sets. Compared with the current four mainstream algorithms (the highest average tracking success rate is 60, the minimum tracking error is 26.14 pixels), the overall performance of this algorithm is optimal. The average tracking success rate is 90, the average tracking error is 7.47 pixels, and the average tracking rate is 25.31 f / s. In the case of target rotation and fast motion, the tracking success rate is 97.4%, and the average relative error of tracking distance is 4.05%.
【作者单位】：华南理工大学机械与汽车工程学院;
【基金】：广东省级科技计划(20148090920001)项目资助
【分类号】：TP391.41

【相似文献】