Visual Tracking & Recognition with Particle Filters
 | Watch the movie (1.3 MB) |
We present a method for simultaneous tracking and recognition of visual objects from video using a time series model with stochastic diffusion. Specifically, by modeling the dynamics with a particle filter we are able to achieve a very stabilized tracker and an accurate recognizer when confronted by pose, scale and illumination variations. We have tested our tracker on real-time face detection (invariant to scale/rotation) as well as tracking vehicles from ground-level views as well as aerial surveillance video of tanks seen from oblique (affine) views.
Background & Objective: Video-based recognition needs to handle uncertainties in both tracking and recognition. We have focused on face recognition for biometric or surveillance applications. We augment a time-series face tracker in the following ways: (i) Modeling the inter-frame motion and appearance changes within the video sequence; (ii) Modeling the appearance changes between the video frames and gallery images by constructing intra- and extra-personal spaces which can be treated as a 'generalized' version of discriminative analysis and (iii) Utilizing the fact that the gallery images are in frontal views
Technical Discussion: Tracking needs modeling inter-frame motion and appearance changes whereas recognition needs modeling appearance changes between frames and gallery images. In conventional tracking algorithms, the appearance model is either fixed or rapidly changing, and the motion model is simply a random walk with fixed noise variance (the number of particles is typically fixed). To stabilize the tracker, we propose the following features: an observation model arising from an adaptive appearance model, an adaptive velocity motions model with adaptive noise variance, and an adaptive number of particles. The adaptive-velocity model is derived using a first-order linear predictor based on the appearance difference between the incoming observation and the previous particle configuration. Occlusion analysis is implemented using robust statistics. Experimental results on tracking visual objects in long outdoor and indoor video sequences demonstrate the effectiveness and robustness of our tracking algorithm. We then perform simultaneous tracking and recognition by embedding them in one particle filter. For recognition purposes, we model the appearance changes between frames and gallery images by constructing the intra- and extra-personal spaces.
| Technical Reports: | |
| Visual Tracking and Recognition Using Appearance-Adaptive Models in Particle Filters | |
| Appearance Tracking Using Adaptive Models in a Particle Filter | |
| Adaptive Visual Tracking and Recognition using Particle Filters | |
Technology Areas:
Computer Vision
Audio Video Processing
Modification Date: September 12, 2007