In order to solve the problem that the feature extraction is easy to be affected by external factors and the computational complexity is high, the depth data was used for human action recognition, which is a more effective solution scheme. Using the joint data collected by Kinect, the human joint was divided into five regions. The vector angle of each region was discretized to describe different states, and then Baum-Welch algorithm was used to study multi-Hidden Markov Model (multi-HMM), meanwhile, forward algorithm was used to establish the generation region and action class probability matrix. On this basis, the region and action categories were intra-coupled and inter-coupled to analyze, thus expressing the interaction between the joints. Finally, the K-Nearest Neighbors (KNN) algorithm based on coupling was used to complete the action recognition. The experimental results show that the recognition rates of the five actions reach above 90%, and the comprehensive recognition rate is higher than that of the contrast methods such as 3D Trajecttories, which means that the proposed algorithm has obvious advantages.

In cognitive satellite terrestrial networks, when the satellite users are secondary users, power control is necessary to guarantee the communication quality of terrestrial primary user in the uplink case. In the context of fading channels, maximizing the Ergodic Capacity (EC) of the satellite user was selected as the objective function, then two optimal power control schemes were proposed based on Peak Interference power Constraint (PIC) and Average Interference power Constraint (AIC), respectively. Meanwhile, the closed expression of the optimal transmit power was given. Simulation results show that the ergodic capacity of satellite user can be increased when the satellite link experiences the weaker shadowing conditions; moreover, under the the specific satellite link condition, the performance of satellite user becomes better with the increasing of the terrestrial interference link fading parameters. In addition, power control method based on AIC is superior to that based on PIC.

In order to improve the accuracy and stability of indoor positioning, an indoor localization algorithm using particle filtering to fuse WiFi fingerprinting and Pedestrian Dead Reckoning (PDR) was proposed. To reduce the negative influence of complex indoor environment on WiFi fingerprinting, a Support Vector Machine (SVM)-based WiFi fingerprinting algorithm using SVM classification and regression for more accurate location estimation was proposed. For smartphone based PDR, in order to reduce the error of inertial sensor, and the effects of random walk, the method of state transition was used to recognize the gait cycles and count the steps, the parameters of state transition were set dynamically using real-time acceleration data, the step length was calculated with Kalman filtering by making use of the relationship between vertical acceleration and step size, and the relationship between adjacent step sizes. The experimental results show that SVM-based WiFi fingerprinting outperformed Nearest Neighbor (NN) algorithm by 34.4% and K-Nearest Neighbors ( KNN) algorithm by 27.7% in average error distance, the enhanced PDR performed better than typical step detection software and step length estimation algorithms. After particle filtering, the trajectory of the fused solution is closer to the real trajectory than WiFi fingerprinting and PDR. The average error distance of linear walking is 1.21 m, better than 3.18 m of WiFi and 2.76 m of PDR; the average error distance of a walking through several rooms is 2.75 m, better than 3.77 m of WiFi and 2.87 m of PDR.

The conventional Multi-dimensional Bayesian Network Classifier (MBNC) requires its structure be bi-partitie. Removing this constraint can result into a new tool named General MBNC (GMBNC), and it enables us to model the underlying joint distribution more correctly. Based on iterative local search of Markov blankets, an algorithm called IPC-GMBNC was proposed to induce the exact structure of GMBNC. The proposed algorithm has good scalability because it does not need to recover the global Bayesian Network (BN) first. The experiments on samples generated from known Bayesian network structures indicate that IPC-GMBNC is effective, and it brings great reduction on computing complexity compared to global search approach, e.g. PC algorithm.