Feature Engineering for Motion Classification in Machine Vision

With the most advanced classification algorithms in the technological platform, the computational power requirement is on the surge. The paper hereby presents computationally trivial algorithms to simplify the process of computational intensive classifications techniques, especially in the Motion Classification arena. The proposed methods prove crucial in acting as a lightweight and computationally fast stepping stone to a fundamentally more significant application of Motion indexing and classification, Action recognition, and predictive analysis of motion energy. The algorithms classify the motions into linear, circular, or periodic motion types by following an appropriate execution order. They consider the tracked motion path of the object of interest as a sequence and use it as a starting point to perform all operations, resulting in a feature that can be classified into separate classes. Using a single parameter for classifying the motion engenders a faster and relatively more straightforward route to motion identification and elicits the algorithm’s uniqueness. Two algorithms are proposed, namely, Angle Derivative Technique and Determinant Method for classifying the motion into two classes (linear & circular). On the other hand, a different algorithm identifies periodic motion using the principle of correlation on the motion sequences. All the algorithms show an average accuracy of over 95%. It also elicited an average processing time of 15.6 ms and 19.86 ms for Angle Derivative Method and Determinant Method, respectively, and 31.2 ms for periodic motion on Intel(R) Core(TM) i3-5005U CPU @ 2.00 GHz and 8GB RAM. A dataset of camera-captured videos consisting of three motion types is used for testing while the proposed methods are trained on a dataset of motion described by mathematical equations with added 3σ noise levels.

THE STUDY IN DIFFUSION MECHANISM BY VORONOI POLYHEDRON IN SODIUM TETRA-SILICATE MELT

Molecular dynamic simulation is carried out for Sodium tetra-silicate (NS4) melt at 1873 K and pressure of 0.1 MPa. The diffusion mechanism of Na atoms is investigated in terms of Voronoi polyhedron around network former and displacement of Na atoms between them. The simulation shows that Na atoms are not uniformly distributed through polyhedrons, but they mainly gather in nonbridging oxygen (NBO) and free oxygen (FO) polyhedrons. More than 75.22% of total Na atoms are place in NBO polyhedrons, although the number of NBO polyhedrons is only 22.27%. The two motion types give mainly contribution to Na diffusion: hopping of isolated Na atom or collective displacement. During 150 ps, the system comprises two separate regions: Na-poor regions formed by Si-O subnets and Na-rich regions formed by O2 clusters. The two regions have strongly different chemical composition, the density of Na atoms as well as motion type of Na atoms.

Evaluating the Accuracy of an Equivalent Linear Model in Predicting Peak Displacement of Seismic Isolation Systems using Single Friction Pendulum Bearings

This study evaluates the accuracy of an equivalent linear model in predicting peak nonlinear time-history displacement of seismic isolation systems with single friction pendulum bearings. To perform this evaluation, dynamic response of numerical models of 120 isolation systems subjected to 390 strong earthquake ground motions, including motions with pulse and motions without pulse, was analyzed and statistically processed. The results show that the equivalent linear model can partly predict the peak displacement of its counterpart nonlinear model. However, the equivalent model can also underestimate or overestimate the peak displacement. On average sense, the equivalent linear model underestimates small peak displacement and overestimates large peak displacement. It is also observed that the relationship between linear and nonlinear peak displacements depends on ground motion types. Based on the analysis data, equations representing relationship between linear and nonlinear peak displacements at different reliable levels for different ground motion types were proposed. These equations can be used in practice.

Real-Time Pedestrian Tracking Terminal Based on Adaptive Zero Velocity Update

In the field of pedestrian dead reckoning (PDR), the zero velocity update (ZUPT) method with an inertial measurement unit (IMU) is a mature technology to calibrate dead reckoning. However, due to the complex walking modes of different individuals, it is essential and challenging to determine the ZUPT conditions, which has a direct and significant influence on the tracking accuracy. In this research, we adopted an adaptive zero velocity update (AZUPT) method based on convolution neural networks to classify the ZUPT conditions. The AZUPT model was robust regardless of the different motion types of various individuals. AZUPT was then implemented on the Zynq-7000 SoC platform to work in real time to validate its computational efficiency and performance superiority. Extensive real-world experiments were conducted by 60 different individuals in three different scenarios. It was demonstrated that the proposed system could work equally well in different environments, making it portable for PDR to be widely performed in various real-world situations.

Upper-Limb Motion Recognition Based on Hybrid Feature Selection: Algorithm Development and Validation (Preprint)

BACKGROUND For rehabilitation training systems, it is essential to automatically record and recognize exercises, especially when more than one type of exercise is performed without a predefined sequence. Most motion recognition methods are based on feature engineering and machine learning algorithms. Time-domain and frequency-domain features are extracted from original time series data collected by sensor nodes. For high-dimensional data, feature selection plays an important role in improving the performance of motion recognition. Existing feature selection methods can be categorized into filter and wrapper methods. Wrapper methods usually achieve better performance than filter methods; however, in most cases, they are computationally intensive, and the feature subset obtained is usually optimized only for the specific learning algorithm. OBJECTIVE This study aimed to provide a feature selection method for motion recognition of upper-limb exercises and improve the recognition performance. METHODS Motion data from 5 types of upper-limb exercises performed by 21 participants were collected by a customized inertial measurement unit (IMU) node. A total of 60 time-domain and frequency-domain features were extracted from the original sensor data. A hybrid feature selection method by combining filter and wrapper methods (FESCOM) was proposed to eliminate irrelevant features for motion recognition of upper-limb exercises. In the filter stage, candidate features were first selected from the original feature set according to the significance for motion recognition. In the wrapper stage, k-nearest neighbors (kNN), Naïve Bayes (NB), and random forest (RF) were evaluated as the wrapping components to further refine the features from the candidate feature set. The performance of the proposed FESCOM method was verified using experiments on motion recognition of upper-limb exercises and compared with the traditional wrapper method. RESULTS Using kNN, NB, and RF as the wrapping components, the classification error rates of the proposed FESCOM method were 1.7%, 8.9%, and 7.4%, respectively, and the feature selection time in each iteration was 13 seconds, 71 seconds, and 541 seconds, respectively. CONCLUSIONS The experimental results demonstrated that, in the case of 5 motion types performed by 21 healthy participants, the proposed FESCOM method using kNN and NB as the wrapping components achieved better recognition performance than the traditional wrapper method. The FESCOM method dramatically reduces the search time in the feature selection process. The results also demonstrated that the optimal number of features depends on the classifier. This approach serves to improve feature selection and classification algorithm selection for upper-limb motion recognition based on wearable sensor data, which can be extended to motion recognition of more motion types and participants.

Predictive encoding of motion begins in the primate retina

ABSTRACTSurvival in the natural environment often relies on an animal’s ability to quickly and accurately predict the trajectories of moving objects. Motion prediction is primarily understood in the context of translational motion, but the environment contains other types of behaviorally salient motion, such as that produced by approaching or receding objects. However, the neural mechanisms that detect and predictively encode these motion types remain unclear. Here, we address these questions in the macaque monkey retina. We report that four of the parallel output pathways in the primate retina encode predictive information about the future trajectory of moving objects. Predictive encoding occurs both for translational motion and for higher-order motion patterns found in natural vision. Further, predictive encoding of these motion types is nearly optimal with transmitted information approaching the theoretical limit imposed by the stimulus itself. These findings argue that natural selection has emphasized encoding of information that is relevant for anticipating future properties of the environment.

Emergent Languages from Pretrained Embeddings Characterize Latent Concepts in Dynamic Imagery

Recent unsupervised learning approaches have explored the feasibility of semantic analysis and interpretation of imagery using Emergent Language (EL) models. As EL requires some form of numerical embedding as input, it remains unclear which type is required in order for the EL to properly capture key semantic concepts associated with a given domain. In this paper, we compare unsupervised and supervised approaches for generating embeddings across two experiments. In Experiment 1, data are produced using a single-agent simulator. In each episode, a goal-driven agent attempts to accomplish a number of tasks in a synthetic cityscape environment which includes houses, banks, theaters and restaurants. In Experiment 2, a comparatively smaller dataset is produced where one or more objects demonstrate various types of physical motion in a 3D simulator environment. We investigate whether EL models generated from embeddings of raw pixel data produce expressions that capture key latent concepts (i.e. an agent’s motivations or physical motion types) in each environment. Our initial experiments show that the supervised learning approaches yield embeddings and EL descriptions that capture meaningful concepts from raw pixel inputs. Alternatively, embeddings from an unsupervised learning approach result in greater ambiguity with respect to latent concepts.

Time analysis of throwing motions to second base by baseball catchers: Comparison of three throwing motions

The purpose of this study was to investigate a baseball catcher’s throwing time to second base using three throwing motion types. The subjects were professional ( n = 4) and college ( n = 12) baseball catchers. Two high-speed cameras were set to capture the throwing motion, while one was set to capture the net on second base. The throwing time of quick throw (throwing motion to release the ball immediately after catching the ball rather than usual throwing motion) was significantly shorter than those of usual throw (throwing motion used during games and practice) and fast ball throw (throwing motion to increase the ball velocity than usual throwing motion). From this result, it became clear that quick throw is the optimal throwing motion when judged by time. Thus, with respect to correlations between variables, there were significant positive correlations between throwing and motion times (usual throw: r = 0.760; fast ball throw: r = 0.719; quick throw: r = 0.767), and between throwing and airborne times (usual throw: r = 0.784; fast ball throw: r = 0.744; quick throw: r = 0.806), for all three throwing motions. However, negative correlations were shown between throwing and release times in usual throw and fast ball throw. The results suggest that, to shorten the throwing time, it is necessary to shorten the hold and stride times and to improve the ability to throw the ball as fast as possible with a shorter motion time.