动态曝光单目视觉惯导系统时间戳延迟估计方法

Timestamp Delay Estimation for a Dynamic Exposure Monocular Visual-inertial System

针对动态曝光单目视觉惯导系统时间戳不同步的问题,提出了一种时间戳延迟估计方法.首先,分析传感器曝光时间、内部数据处理、滤波和数据传输等不同因素产生的时间延迟,并在此基础上建立动态曝光时间戳延迟模型;其次,基于B-样条曲线拟合和互相关方法得到延迟时间的粗略估计,然后使用非线性优化方法对构造的目标函数进行优化以得到不同曝光时间对应的时间戳延迟精确估计;最后,基于最小二乘方法,使用所建立的延迟模型对上一步的延迟估计结果进行拟合得到模型的参数估计.实际物理实验结果表明,所提出的动态曝光时间戳延迟模型与实际物理硬件延迟特性相符,并且通过所提出模型对时间戳延迟进行估计和补偿后可以提高视觉惯性里程计的估计精度,从而为基于视觉的定位、建图、导航及路径规划提供更精确的数据支持.

Aiming at solving the problem of timestamp misalignment of dynamic exposure monocular visual-inertial system (VINS), a timestamp delay estimation method was proposed. First, different factors, which lead to the timestamp delay, such as sensor exposure time, internal sensor latency, filtering and data transmission, were analyzed. With the above analysis, the dynamic exposure timestamp delay model was established. Second, based on the B-spline curve fitting and cross-correlation, a rough estimation of the time delay was obtained. Then, the objective function was constructed and optimized by nonlinear optimization method to obtain the accurate estimation of the timestamp delays corresponding to different exposure times. Finally, the parameters of the dynamic exposure timestamp delay model were obtained by fitting the estimated results to the proposed model with least squares. The real physical experimental results show that the proposed dynamic exposure timestamp delay model is consistent with the actual physical hardware time delay characteristic, and the estimation accuracy of visual-inertial odometry is improved after compensating the timestamp delay with the proposed model, which can provide more accurate data for vision-based localization, navigation, mapping and path planning.