论文部分内容阅读
针对现有跟踪算法不能很好地适应强非线性动态跟踪问题,提出利用稀疏特性更强的相关向量机(RVM)对换道参数进行估计,并采用样条曲线对估计值进行规划。将车辆纵向车速和横向车速以及横摆角速度作为表征参数,依据真车试验数据建立相应的运动方程、RVM输出方程以及条件概率方程,选取具有良好非线性特性的高斯函数作为核函数,依据不同带宽下模型性能的优劣确定最佳的带宽。对较为敏感的横摆角速度进行了滤波处理,通过对边缘似然函数进行迭代求解确定出权重分布和估计噪声,并采用RVM模型对换道参数进行估计,利用B样条曲线规划换道轨迹。测试结果表明:RVM具有良好的估计特性,相比SVM而言,其对核函数的敏感度较低,测试时间短,对大样本数据具有良好的适应特性,经过B样条曲线规划后,估计值的连续性和尖峰特性得到最大限度地改善。
Aiming at the problem that the existing tracking algorithms can not well adapt to the strong nonlinear dynamic tracking problem, a new correlation vector machine (RVM) with more sparse features is proposed to estimate the lane parameters and the spline curve is used to estimate the estimated values. According to the real car test data, the corresponding motion equations, RVM output equations and conditional probability equations are established. The Gaussian function with good nonlinearity is selected as the kernel function, and the bandwidth is calculated according to different bandwidth The performance of the next model determines the best bandwidth. The more sensitive yaw rate is filtered, and the weight distribution and estimated noise are determined by iteratively solving the edge likelihood function. The RVM model is used to estimate the lane changing parameters, and the lane-changing trajectories are planned by B-spline curves. The test results show that the RVM has good estimation performance. Compared with SVM, the RVM has lower sensitivity to kernel function, shorter test time and good adaptability to large sample data. After B-spline curve programming, RVM estimates The continuity and spike characteristics of values are maximized.