一种基于速度预测的双目视觉跟踪混合算法

A Binocular Vision Tracking Hybrid Algorithm Based on Speed Prediction

视觉跟踪技术作为计算机视觉研究的重要方向,得到了广泛应用。其根据目标表征方式可分为生成式方法和判别式方法。传统生成式方法虽然有较好的实时性,但容易受到复杂背景和遮挡影响,所以逐步被判别式方法所取代。针对传统生成式方法的缺点,结合双目成像系统空间定位的优势,提出了一种基于速度预测的双目视觉跟踪混合算法。通过帧间差分法发现视野中运动目标,采用双目定位确定其运动速度,并根据速度预估物体的下一帧位置;同时,使用ORB特征匹配、双目视觉极线约束和Mean-Shift进行验证和速度修正。试验证明,该算法能够准确地识别运动物体并对其进行跟踪,克服了生成式方法容易受到环境干扰的缺点,保留了其实时性的优势,具有较高的实用价值。

As an important direction of computer vision research,the visual tracking technology is widely used.In accordance with the target characterization,it can be divided into generative method and discriminant method.Although traditional generative method has better realtime performance,it is easily influenced by complicated background and occlusion,so it is gradually replaced by discriminant method.Aiming at the shortcomings of traditional generative method and combining the superiority of binocular imaging system in spatial positioning,a binocular vision tracking hybrid algorithm based on speed prediction is proposed.The moving target in the view is found through the interframe differential method,and its moving speed is determined by binocular positioning.The object’s position in next frame is estimated by the speed,and the ORB feature matching,binocular polar constraint and MeanShift are used to verify and correct the speed.Experiments prove that this algorithm can accurately recognize moving objects and track them,overcome the shortcoming of generative methods which are easily disturbed by the environmental interference,while retains the advantages of realtime performance,so it has high practical value.