Francis Li

PIRM: Fast background subtraction under sudden, local illumination changes via probabilistic illumination range modelling

We present an illumination-compensation method to enable fast and reliable background subtraction under sudden, local illumination changes in wide area surveillance videos. We use Probabilistic Illumination Range Modeling (PIRM) to model the conditional probability distribution of current frame intensity given background intensity. With this model, we can identify a continuous range of current frame intensities that map to the same background intensity, and scale all pixels within that range in the current frame appropriately to enable illumination-compensated background subtraction. Experimental results using a standard academic dataset as well as very challenging industry videos show that PIRM can achieve improvements in compensating for sudden, local illumination changes.

Simultaneous Projector-Camera Self-Calibration for Three-Dimensional Reconstruction and Projection Mapping

Automatic calibration of structured-light systems, generally consisting of a projector and camera, is of great importance for a variety of practical applications. We propose a novel optimization approach for geometric calibration of a projector-camera system that estimates the intrinsic, extrinsic, and distortion parameters of both the camera and projector in an automatic fashion using structured light. Our approach benefits from a novel multifactor objective function that finds maximum-likelihood estimates from noisy point correspondences using constraints on focal lengths and resolves ambiguities estimating the fundamental matrix by enforcing epipolar geometry on the rectified noisy data. This new formulation allows estimation of all calibration parameters simultaneously and minimization is ensured by a greedy descent algorithm that decreases the cost function at each iteration. This provides more accurate parameter estimation, reconstruction accuracy, and robustness to noise and poor initialization compared to previous methods. Experimental results demonstrate the stability and robustness of our method, and show that the proposed solution outperforms a currently leading approach to an automatic geometric projector-camera calibration.

Dense Depth Map Reconstruction from Sparse Measurements Using a Multilayer Conditional Random Field Model

Acquiring accurate dense depth maps is crucial for accurate 3D reconstruction. Current high quality depth sensors capable of generating dense depth maps are expensive and bulky, while compact low-cost sensors can only reliably generate sparse depth measurements. We propose a novel multilayer conditional random field (MCRF) approach to reconstruct a dense depth map of a target scene given the sparse depth measurements and corresponding photographic measurements obtained from stereo photogrammetric systems. Estimating the dense depth map is formulated as a maximum posterior (MAP) inference problem where a smoothness prior is assumed. Our MCRF model uses the sparse depth measurement as an additional observation layer and describes relations between nodes with multivariate feature functions based on the depth and photographic measurements. The method is first qualitatively analyzed when performed on data collected with a compact stereo camera, then quantitative performance is measured using the Middlebury stereo vision data for ground truth. Experimental results show our method performs well for reconstructing simple scenes and has lower mean squared error compared to other dense depth map reconstruction methods.

Hierarchical Grouping Approach for Fast Approximate RGB-D Scene Flow

A new approach to efficiently compute RGB-Dscene flow is introduced based on matching 3D points fromone frame to the next in a hierarchical fashion. Most stateof-the-art RGB-D scene flow methods are set in a variationalframework and formulated as an energy minimization problem. While these methods are able to provide high accuracy, theyare computationally expensive and not robust under largermotions in the scene. As well, the RGB-D scene flow datasetspresented to date are mostly based on qualitative evaluationof real scenes. The main contributions of this work are topresent a method of efficiently computing approximate sceneflow and provide an RGB-D scene flow dataset with groundtruth flow for quantitative evaluation. Quickly determiningapproximate motions in a scene is tremendously useful forany computer vision tasks that benefit from motion cues suchobstacle avoidance, object recognition, action recognition, etc. The proposed method, named Hierarchical Spectral GroupingScene Flow (HSG-SF), uses a simple coarse-to-fine voxelizationscheme combined with spectral grouping methods to providefast estimates of motion and accommodate for larger motions. Experimental results show that HSG-SF can provide reliablescene flow estimates at significantly faster runtime speed thancurrent methods.