The artificial potential field approach provides a simple and effective motion planner for robot navigation. However, the traditional artificial potential field approach in practice can have a local minimum problem, i.e., the attractive force from the target position is in the balance with the repulsive force from the obstacle, such that the robot cannot escape from this situation and reach the target. Moreover, the moving object detection and avoidance is still a challenging problem with the current artificial potential field method. In this paper, we present an improved version of the artificial potential field method, which uses a dynamic window approach to solve the local minimum problem and define a danger index in the speed field for moving object avoidance. The new danger index considers not only the relative distance between the robot and the obstacle, but also the relative velocity according to the motion of the moving objects. In this way, the robot can find an optimized path to avoid local minimum and moving obstacles, which is proved by our experimental results.

Distributed camera network for human pose estimation can solve the problem of limited view and occlusion of single view, which has great potential for wide area surveillance applications. To fuse different field of view information, we propose a distributed human pose estimation method by combining the interactive multiple model (IMM) algorithm with the distributed information fusion of human skeleton joints. Compared to the state of the art works which often use a single motion model to depict the motion of human skeleton joints, e.g., constant velocity, the novelty of our work is that the maneuvering property of human action is handled by IMM, i.e., the motion model of human skeleton joints in the filter is approximated using constant velocity, constant acceleration and Singer motion models. Based on the advantages of IMM algorithm for maneuvering target tracking, our method can not only solve the single-view occlusion problem, but also solve the problem of joint point fluctuation caused by the estimation error of each sensor node after distributed information fusion. The final human action recognition experimental results show that the proposed method can improve the action recognition rate on the datasets captured by Kinect V2. In addition, we built a distributed camera network using embedded machine learning boards, such that deep learning based human pose estimation methods can be employed in our framework to handle the limitations of original Kinect SDK.

In this paper, we introduce the information weighted consensus filter (IWCF) based human pose estimation method for action recognition using a distributed RGBD camera network, which can handle the occlusion problem effectively existed in the single view image. As far as we know, this is the first time that the novel consensus filter is proposed for distributed 3D human pose estimation and action recognition. To demonstrate the performance of the proposed idea, we construct a small RGBD camera network, and collect a number of actions from different persons. The final recognition results show that the IWCF based human pose estimation method can converge to the centralized solution and achieve 93.75% accuracy rate after 9 consensus iterations, which is much higher than the recognition results derived from single views.

This paper presents a probabilistic algorithm for lane shape estimation in an urban environment which is important for example for driver assistance systems and autonomous driving. For the first time, we bring together the so-called Partitioned Particle filter, an improvement of the traditional Particle filter, and the linear-parabolic lane model which alleviates many shortcomings of traditional lane models. The former improves the traditional Particle filter by subdividing the whole state space of particles into several subspaces and estimating those subspaces in a hierarchical structure, such that the number of particles for each subspace is flexible and the robustness of the whole system is increased. Furthermore, we introduce a new statistical observation model, an important part of the Particle filter, where we use multi kernel density to model the probability distribution of lane parameters. Our observation model considers not only color and position information as image cues, but also the image gradient. Our experimental results illustrate the robustness and efficiency of our algorithm even when confronted with challenging scenes.