Multi-Resolution Estimation of Optical Flow for Vehicle Tracking

This paper presents a hierarchical multi-resolution estimation of optical flow for a vehicle tracking system which can be used in a practical environment. Aiming at accurate estimation of optical flow, we construct a strong feature tracking system based on the Shi-Tomasi approach. As a feature detector, we use a Scale-Invariant Feature Transform (SIFT) algorithm, which not only firmly focuses on multi-scaling images but also correctly tracks strong interest points. The pyramidal Lucas-Kanade optical flow algorithm using our feature tracking system is then implemented. The information after estimating the optical flow can be used in the later tracking and detecting processes. For evaluation, we use the Autonomous Agents for On-Scene Networked Incident Management (ATON) project’s highway video files, which include moving shadows. To test the tracking in a practical environment, we artificially add three adverse effects: additive Gaussian noise, vibration and motion blur. Experimental results demonstrate good performance for our multi-resolution optical flow system. The study confirms that the computed optical flow has a very small error with minimum eigenvalues in the optical flow equations.