Johnny Park

Distributed Object Tracking Using a Cluster-Based Kalman Filter in Wireless Camera Networks

Local data aggregation is an effective means to save sensor node energy and prolong the lifespan of wireless sensor networks. However, when a sensor network is used to track moving objects, the task of local data aggregation in the network presents a new set of challenges, such as the necessity to estimate, usually in real time, the constantly changing state of the target based on information acquired by the nodes at different time instants. To address these issues, we propose a distributed object tracking system which employs a cluster-based Kalman filter in a network of wireless cameras. When a target is detected, cameras that can observe the same target interact with one another to form a cluster and elect a cluster head. Local measurements of the target acquired by members of the cluster are sent to the cluster head, which then estimates the target position via Kalman filtering and periodically transmits this information to a base station. The underlying clustering protocol allows the current state and uncertainty of the target position to be easily handed off among clusters as the object is being tracked. This allows Kalman filter-based object tracking to be carried out in a distributed manner. An extended Kalman filter is necessary since measurements acquired by the cameras are related to the actual position of the target by nonlinear transformations. In addition, in order to take into consideration the time uncertainty in the measurements acquired by the different cameras, it is necessary to introduce nonlinearity in the system dynamics. Our object tracking protocol requires the transmission of significantly fewer messages than a centralized tracker that naively transmits all of the local measurements to the base station. It is also more accurate than a decentralized tracker that employs linear interpolation for local data aggregation. Besides, the protocol is able to perform real-time estimation because our implementation takes into consideration the sparsity of the matrices involved in the problem. The experimental results show that our distributed object tracking protocol is able to achieve tracking accuracy comparable to the centralized tracking method, while requiring a significantly smaller number of message transmissions in the network.

A probabilistic framework for joint segmentation and tracking

Most tracking algorithms implicitly apply a coarse segmentation of each target object using a simple mask such as a rectangle or an ellipse. Although convenient, such coarse segmentation results in several problems in tracking — drift, switching of targets, poor target localization, to name a few — since it inherently includes extra non-target pixels if the mask is larger than the target or excludes some portion of target pixels if the mask is smaller than the target. In this paper, we propose a novel probabilistic framework for jointly solving segmentation and tracking. Starting from a joint Gaussian distribution over all the pixels, candidate target locations are evaluated by first computing a pixel-level segmentation and then explicitly including this segmentation in the probability model. The segmentation is also used to incrementally update the probability model based on a modified probabilistic principal component analysis (PPCA). Our experimental results show that the proposed method of explicitly considering pixellevel segmentation as a part of solving the tracking problem significantly improves the robustness and performance of tracking compared to other state-of-the-art trackers, particularly for tracking multiple overlapping targets.

3D Modeling of Optically Challenging Objects

We present a system for constructing 3D models of real-world objects with optically challenging surfaces. The system utilizes a new range imaging concept called multipeak range imaging, which stores multiple candidates of range measurements for each point on the object surface. The multiple measurements include the erroneous range data caused by various surface properties that are not ideal for structured-light range sensing. False measurements generated by spurious reflections are eliminated by applying a series of constraint tests. The constraint tests based on local surface and local sensor visibility are applied first to individual range images. The constraint tests based on global consistency of coordinates and visibility are then applied to all range images acquired from different viewpoints. We show the effectiveness of our method by constructing 3D models of five different optically challenging objects. To evaluate the performance of the constraint tests and to examine the effects of the parameters used in the constraint tests, we acquired the ground-truth data by painting those objects to suppress the surface-related properties that cause difficulties in range sensing. Experimental results indicate that our method significantly improves upon the traditional methods for constructing reliable 3D models of optically challenging objects.

Hierarchical Data Structure for Real-Time Background Subtraction

This paper seeks to increase the efficiency of background subtraction algorithms for motion detection. Our method uses a quadtree-base hierarchical framework that samples a small portion of the pixels in each image and yet produces motion detection results that are very similar compared to the conventional methods that raster scan entire images. The hierarchical data structure presented in this paper can be used with any background subtraction algorithm that employs background modeling and motion detection on a per-pixel basis. We have tested our method using two common background subtraction algorithms: running average and mixture of Gaussian. Our experimental results show that the application of the hierarchical data structure significantly increases the processing speed for accurate motion detection. For example, the mixture of Gaussian method with our hierarchical data structure is able to process 1600 by 1200 images at 11~12 frames per second compared to 2~3 frames per second without using the hierarchical data structure.

A Light-Weight Event-Driven Protocol for Sensor Clustering in Wireless Camera Networks

We propose a light-weight event-driven protocol for wireless camera networks to allow for formation and propagation of clusters of cameras for the purpose of collaborative processing during object tracking. Cluster formation is triggered by the detection of objects with specific features. Our protocol allows for simultaneous formation and propagation of multiple clusters. Cameras being directional devices, more than one cluster may track a single object since groups of cameras outside each others communication range may see the same object. Entry into a cluster and cluster membership maintenance require a sensor node to confirm the presence of features of the object being tracked. Each cluster elects its own leader among the cameras that observe the same target. When a cluster leader loses track of an object, it assigns the leadership role to another cluster member. To avoid high communication overhead among cluster members, single-hop clusters are formed, i.e., every member of a cluster is within the communication range of the cluster head. We have implemented a simple version of this protocol on a test-bed and provide an experimental evaluation.

Segmentation of Apple Fruit from Video via Background Modeling

A method for locating apples was developed to process real-time video image sequences captured with an over-the-row harvester. The concepts of background modeling in RGB color were used, which is a novel approach to the apple segmentation problem. In background modeling, the distributions of background colors are approximated from real data. The algorithm developed for this task, Global Mixture of Gaussians (GMOG), is based on the principles of Mixture of Gaussians (MOG), which is used for motion-detection applications. The algorithm correctly identified ~85-96% of both red and yellow apples and performed at ~14-16 frames per second. This is the first work to our knowledge that uses video sequences to detect apple fruit. The potential advantages of using video include allowing harvesting on-the-go, detecting occluded fruit via camera movement to the occluded areas, using visual servoing of robotic grippers to grasp fruit, and achieving a faster harvest time.

A parallel histogram-based particle filter for object tracking on SIMD-based smart cameras

We present a parallel implementation of a histogram-based particle filter for object tracking on smart cameras based on SIMD processors. We specifically focus on parallel computation of the particle weights and parallel construction of the feature histograms since these are the major bottlenecks in standard implementations of histogram-based particle filters. The proposed algorithm can be applied with any histogram-based feature sets-we show in detail how the parallel particle filter can employ simple color histograms as well as more complex histograms of oriented gradients (HOG). The algorithm was successfully implemented on an SIMD processor and performs robust object tracking at up to 30 frames per second-a performance difficult to achieve even on a modern desktop computer.

A parallel color-based particle filter for object tracking

Porting well known computer vision algorithms to low power, high performance computing devices such as SIMD linear processor arrays can be a challenging task. One especially useful such algorithm is the color-based particle filter, which has been applied successfully by many research groups to the problem of tracking non-rigid objects. In this paper, we propose an implementation of the color-based particle filter suitable for SIMD processors. The main focus of our work is on the parallel computation of the particle weights. This step is the major bottleneck of standard implementations of the color-based particle filter since it requires the knowledge of the histograms of the regions surrounding each hypothesized target position. We expect this approach to perform faster in an SIMD processor than an implementation in a standard desktop computer even running at much lower clock speeds.

Comprehensive Automation for Specialty Crops: Year 1 results and lessons learned

Comprehensive Automation for Specialty Crops is a project focused on the needs of the specialty crops sector, with a focus on apples and nursery trees. The project’s main thrusts are the integration of robotics technology and plant science; understanding and overcoming socio-economic barriers to technology adoption; and making the results available to growers and stakeholders through a nationwide outreach program. In this article, we present the results obtained and lessons learned in the first year of the project with a reconfigurable mobility infrastructure for autonomous farm driving. We then present sensor systems developed to enable three real-world agricultural applications—insect monitoring, crop load scouting, and caliper measurement—and discuss how they can be deployed autonomously to yield increased production efficiency and reduced labor costs.

Distributed and lightweight multi-camera human activity classification

We propose a human activity classification algorithm that has a distributed and lightweight implementation appropriate for wireless camera networks. With input from multiple cameras, our algorithm achieves invariance to the orientation of the actor and to the camera viewpoint. We conceptually describe how the algorithm can be implemented on a distributed architecture, obviating the need for centralized processing of the entire multi-camera data. The lightweight implementation is made possible by the very affordable memory and communication bandwidth requirements of the algorithm. Notwithstanding its lightweight nature, the performance of the algorithm is comparable to that of the earlier multi-camera approaches that are based on computationally expensive 3D human model construction, silhouette matching using reprojected 2D views, and so on. Our algorithm is based on multi-view spatio-temporal histogram features obtained directly from acquired images; no background subtraction is required. Results are analyzed for two publicly available multi-camera multi-action datasets. The systems advantages relative to single camera techniques are also discussed.