 Research
 Open Access
 Published:
A hybrid approach to inferring the Internet of Things for complex activity recognition
EURASIP Journal on Wireless Communications and Networking volume 2019, Article number: 251 (2019)
Abstract
With the rapid development and largescale uptake of the Internet of Things, smart home is evolving from a vision towards a realistically viable solution for assisted living. Activity recognition is one of the fundamental tasks in order to provide accurate and timely assistance and service. As daily living scenarios are full of similar activities, missing data, and noise, inferring complex activities using knowledgedriven reasoning algorithms suffers from several drawbacks, e.g., realtime raw sensor data segmentation, poor generalization, higher computational complexity, and scalability. To address these problems, this paper proposes a hybrid approach to complex daily activity recognition by merging the firstorder logic and probability graphic modeling. Specifically, we develop a novel “Markov logic network” combining datadriven multifeature and simplified rulebased modeling and inference, thus enabling and supporting the applicability and robustness of daily activity recognition. To evaluate the approach and associated methods, we design a testing scenario with a number of similar activity groups, missing data, or disturbance test datasets in a multimodeling sensor scene. Initial results show our approach outperforms the traditional approach with a better accuracy in the situations of similar activities with missing data and noise disturbance. Experiments are also conducted to compare the Gibbs sampling and MCSAT sampling algorithms for Markov logic network, and the results show that the Gibbs is better in our experimental settings.
Introduction
Activity recognition is a fundamental task for pervasive computing, contextaware computing, smart activity, healthy aging, wise information technology, and etc. [1]. In the literature of activity recognition, most of the works build activity model based on the temporal spatio characters [2, 3], which faces the miscalculation with the similar activities because of the few features. Another research bottleneck is the poor robustness of the activity model which is difficult to handle the data missing, noisy data, habit changing, and so on [4, 5]. In addition, the complex multiactivities are more popular which require an active. Therefore, designing a comprehensive solution for the daily activity recognition is the main work of this paper.
Recognition of activity of daily living (ADL) aims to recognize the residents’ action intent or intention (activity), living habits, and health status to detect the abnormal behavior and provide the personalized service. There are three types of devices to collect data: vision equipment, wearable devices [9], and ambient sensors. The wearable devices need to be worn all time and charged in time; the wearable devices are excluded in our research which sometimes cause interference. Vision and ambient sensors have their own merits and demerits respectively. The vision sensors have been widely adopted directly [10, 11] which is obvious to record in video. While the privacy requirements have been paid attention to gradually, ambient sensors are easier to be accepted by residents. Therefore, the nonvision ambient sensors have become the research hotspot [12]. That method also has another advantage that the multimodeling sensing data has more proofs for the inference to make the more accurate result. Like human beings, the feeling of touching, smelling, tasting, and hearing and their attributes can be sensed and visible by sensors. For many home activities, ambient sensors like infrared, touch, pressure, and light are more intuitive to represent their related features. A typical indoor scene layout of multimodeling sensors has been shown in Figs. 1 (kitchen), 2 (living room), 3 (water machine), 4 (cup model), and 5 (deployed cup model).
Raw data collection is the first step, and the second step is establishing the inference model. The usual inference methods adopted can be divided two categories, datadriven and knowledgedriven measures. The datadriven method trains the activity model which represents the implicit general rules by huge numbers of labeled data. Actually, the earliest studies have adopted the datadriven method mostly [13–15]. This method is a typical probability statistical model which has good performance on single resident’s single activity recognition. For example, the wellknown key technologies include hidden Markov models and dynamic Bayesian network. It belongs to the series probability graphical model which has the high accuracy and the ability of mining complex correlation for complex activities. But the model is usually trained by the experimenters who are not the users. There has an essence problem for the different distribution between test data and train data which means the model is not impracticable and inapplicable to recognize the personalized activities. Another method, knowledgedriven, is an expert system relying on the background and domain knowledge. This knowledge builds mapping from activity entity, appliance entity and context (sensor entity, sensor value entity), temporalspatial traits, hierarchical methodology, and rules (formulas) [16, 17]. This method has shown a good performance in using directly to overcome “cold start” problem [18]. The general rules of the model represent most residents’ activity habits which have a high reusability, while the preference, details, and sufficient description of one activity cannot be taken into full consideration. For a better performance, the activity model is dynamic and uncertain which should be updated with the habits changing. How to match best results? Try to combine the two methods to learn from other’s strong points to offset one’s weakness which is called the hybrid model [19]. Gabriele proposed a novel hybrid approach with probabilistic and knowledgebased reasoning and adopted the unsupervised feedback to label the rules which have good performance in the realworld dataset [20]. Shin et al. presented the regular expressionbased string matching algorithm; the pattern model is trained by labeled data based on the experts’ knowledge base [21]. These hybrid methods have good performance for complementing and correcting each other. Expert knowledge is more consistent which is utilized at the first step to model the basic formula [22]. Data modification is more flexible which revises the formula. In this paper, we adopt one hybrid model, Markov logic network (MLN), which uses both of the firstorder logic (FOL) and statistical probability. Establish the rules’ model by expert knowledge, then give the satisfiability probability to every rule by learning from series of real raw sensor data sequences. The main advantages are the the strong ability for complex activities and having the good performance to handle the dynamic and the data missing. The details are shown in Section 3.
The main contributions of our paper are as follows:
Improving the Markov logic network by combining the simple words which can express the action, entity, time, time period, and place, which covers all attributes of sensor data and reduces the description complexity
In order to improve the robustness of the activity model, we adopt the subrules to express the activity which just has one evidence for the result. Inference is a combination of the rules which is related to the activity
Proving the Gibbs sampling algorithm has a better performance than the traditional adopted method, MCSAT algorithm, in these complex activities
In this paper, Section 2 presents the background, Markov logic network, which merges the firstorder logic and Markov network. A novel method based on the Markov logic network has been shown in Section 3, which includes data expression, entity expression, activity expression with time series and periods, and simplifying rule expression. Section 4 describes the experiment design and settings. Section 5 shows the results and discussion of different complex activity situations and contrasts the different algorithms in these special situations. Section 6 gives the conclusion and perspective.
Background (MLN)
Graphical model of the firstorder logic
The firstorder logic formula, also called as the firstorder predicate calculus, consists of assertion and quantify. Constant symbols (e.g., time: 20181223 and 20190107) represent the real instance. Variable symbols (e.g., x and y) are the abstraction of constant symbols which range over the real instance. Predicate symbols (e.g., m UseBootle, UseTeabag) represent the relations and attributes mapping into a set of category events, including sensor events, entity events, and activity events. In addition, the object has its own attributes like location and usage which is one feature value to be added into layers. Term is any expression of object (constant, variable) (e.g., 20181223, x), like sensor or entity events, with the time character, for example, dayhourminutesecond. Atom is a predicate symbol applied to a tuple of terms or one term (e.g., UseTeabag(b)). Formulas are recursively constructed from atom using logical connectives and quantifiers, for example, negation (¬), conjunction (∧), disjunction (∨), implication (→), and equivalence (⇔) [23]. It is convenient to convert formulas to a more regular form, called clausal form which is defined by disjunctive normal form (DNF) [24]. Knowledge base is a conjunction of clauses, every clause is a disjunction of literals.
Graph G is composed by nodes and edges: nodes represent the terms, edges represent the relations, and clique of G represents firstorder logic formulas (atoms). The formula is nonnegative and realvalued. According to the graphical methodology, there is an edge between two nodes that appear in at least one formula.
Learning weight of the firstorder logic by probability
Firstorder logic formula weights can be learned generatively by maximizing the likelihood of a relational training database. Automatically giving weight by data is much better than manually refining with less work. The learning algorithm is based on convex optimization. The gradient descent algorithm is one of the optimization methods for searching in firstorder class which use the gradient g, with learning rate η to update the weight ω; the formula has been shown in (1) [25]. The formula of g has been shown in (2), where E_{ω,y} is the expectation over the nonevidence atoms. For the secondorder class, the optimization method aims at searching direction from function as a quadratic surface [26].
A typical methodbased firstorder logic and probability (MLN)
Markov logic network (MLN) is a combination of the firstorder logic and Markov networks [27]. Its knowledge model is firstorder formulas with weights. The Markov network is one of the models for the joint distribution of a set of variables which is an undirected graph G and a set of potential function ϕ_{k}. Each state of M_{L,C} (all nodes have values) represents a possible world.
The conception of MLN, L is a set of pairs (F_{i},ω_{i}), with a finite set of constants C={c_{1},c_{2},…,c_{C}}, defines a M_{L,C} (Markov network). In M_{L,C}, the nodes are binary value, 1 if the ground predicate is true, 0 otherwise. In M_{L,C}, each possible grounding of each formula has been represented. The fewer formulas a world violates, the more probable it is. The associated weight is bigger, reflecting the constraint is stronger.
The probability distribution over possible world x specified by the ground Markov network M_{L,C} is given by formula (3) [27], where n_{i}(x) is the number of the true groundings of F_{i} in x, x_{{i}} is the state (truth values) of the predicates appearing in F_{i}, and \(\phantom {\dot {i}\!}\phi _{i}(x_{\{i\}}) = e^{\omega _{i}}\) [28]. The second right part is the most convenient approach with a mixture of hard and soft constraints. Because for some formulas with 0 potential, it means the probability is 0.
Reasoning probabilities for complex relationships are the key step of this paper. Due to the size and complexity of the grounding Markov network, it is infeasible to inference many scenarios [29]. MLN has two inference types, one is the most likely state, another is the conditional probabilities. We adopt the second inference to calculate every ground query atom’s conditional probabilities. Gibbs sampling is a typical algorithm of the Markov Chain Monte Carlo algorithm. The MCSAT algorithm is a slice sampling Markov Chain Monte Carlo algorithm which combines the satisfiability testing and simulated annealing. Satisfiability solver (WalkSAT) is the efficiently finding isolated modes in the distribution [30].
All relevant atoms have been retrieved by the Markov blanket concept which has been mentioned before. The probability of a ground predicated X_{l} when its Markov blanket B_{l} is in state b_{l} is in (4). F_{l} is the set of ground formulas that X_{l} appears in, and f_{i}(X_{l}=x_{l},B_{l}=b_{l}) is the value of the ith ground formula when X_{l}=x_{l} and B_{l}=b_{l}.
MLN has many outstanding advantages in the following:
Merging multiple expert knowledge even though these formulas are exclusive and inconsistent
Learning and inferencing by ground atoms, even the not independent and identically distributed atoms
Reducing the reliance on expert knowledge
Method
This paper offers a solution method based on MLN to build a recognition model from the bottom up, including the raw data preprocessing, expert system for knowledge rules, rules’ weight learning by labeled data, and inference. The flow chart of the novel method has been shown in Fig. 6.
In Section 1, the processing approach for sensor data has been mentioned. In this section, the hierarchical concept graph will be explained in details in Fig. 7.
There are three layers, the first layer is converting the raw sensor data to an entity event. Sensors have been attached in physical objects and reflect an attribute of this object [31]. The entity event layer is established according to the activity concept. Activity is a series of entity sequences by single or multiple residents. An entity event is one action which has been triggered by a series sensor because the action must have closed connection with one physical entity.
Expression of sensor data
The collecting data of sensors has been stored as the following format. A data record has the sensor name, sensor id, sensor value, sensor triggered time, and sensor place.
In order to recognize the similar activities, add the time period feature and time series to the data. Combine all attributes in one compact expression like:
The specific definition of the sensor, place, and time period has been shown in Table 1, time including the year, month, day, hour, minute, and second.
Simplifying the activity rules by entity expression
The typical entity event is action events which consist of several continuous sensor data; multiple sensor events are triggered in a period of time. The related sensors can be combined together to generate the entity event which can be divided to three categories.
Motion, touch, tilt, pressure,and light
Action is one kind of the entity event, which leads to the position change or the relative position change of an entity. Using multiple kinds of sensors, including the human sensor and object sensor, closing to the object combining with the object selfchange is the entity event which is triggered by a resident. The comprehensive rules of an entity event just concern a very short time of just several seconds and require the small range space to avoid a misunderstanding of the multiresident.
Pressure
Weight change is also one kind of the entity event, which has the significant show for the drink, eat, and other weight transfer events obviously.
Gas, temperature, and humidity
Environment change is key to the context awareness which is one kind of the entity event, when the environment parameter crosses the threshold value; the sensing data has been recorded which can provide assistance for other change situations.
Note that the time span of the entity event is just seconds which can be ignored, reducing the demand for storage and simplifying the inference process with the simple expression of evidence.
The entity event expression is shown in the following:
Activity expression
Activity events have more attributes than sensor and entity events, which add time period and change the triggered time to begin time and end time which is accurate to express the activities. The time period is assisting to distinguish similar activities which have similar entity event series. The evaluation and experiment have been shown in Section 4.
The activity event expression is shown in the following:
Time series expression
The time relationship has three kinds, Before, After, and Equal. There are some definitions of these relations.
Before(x,y)⇔After(y,x)
Before(x,y)∨Equal(x,y)⇔¬Before(y,x)
After(x,y)∨Equal(x,y)⇔¬After(y,x)
Equal(x,y)⇔Equal(y,x)
In the time relationship between the entity event (entity_time) and the activity event (activity_begintime, activity_endtime), we define that the activity begin time and end time have the relation as Before(activity_begintime,activity_endtime). There were five kinds which are incompatible between each other, Before, After, Between, Begin, and End. The specific definition of them is shown in the following:
Before(entity_time,activity_begintime)⇔Before(entity,activity)
After(entity_time,activity_endtime)⇔After(entity,activity)
After(entity_time,activity_begintime)∧Before(entity_time,activity_endtime)⇔Between(entity,activity)
Equal(entity_time,activity_begintime)⇔Begin(entity,activity)
Equal(entity_time,activity_endtime)⇔End(entity,activity)
Simplifying rule expression (subrules)
References [32] and [33] have adopted MLN to recognize the activities, but they work mainly for sequence activities without the probability learning which is just the knowledgedriven method. Civitarese et al. have presented the duration concept by calculating the difference between the beginning time and ending time [20]. These papers design the multirestricted conditions in one rule, but the satisfiability probability is for the whole rule which let the rejection for the rule just by a small dissatisfaction condition. But the dissatisfaction condition is generated by many situations, including the data missing, dynamic habits, and noisy data. In order to improve the flexibility of the activity model, the clausal form just includes the two event atoms which are group subrules of the traditional rules, one is an entity event, one is an activity event. While the time series of entity events also is one of the key evidence for inference, the time relationship between the entity event and activity event is also contained into the clausal.
For example, before we define the Drink activity rules as UseCup(a)_(x,z,p)∧DecreaseCup(a)_(y,z,p)→Drink_(z,p,x,y), now we divide the rule to subrules as UseCup(a)_(x,z,p)→Drink_(z,p,x,y) and UseCup(a)_(x,z,p)→Drink_(z,p,x,y). The two evidence entity events have been consisted into different rules which can improve the flexibility of the inference.
Experiments
We designed three similar group (Drink Tea, Drink Coffee, Have Meal, Do Dishes, Sweep, Wipe) activities in kitchen and living room by a topdown approach in Table 2. In our experiment, two volunteers have been living in the environment 2 weeks (as the small training data set to improve the model), and label all the training data. Then, we have collected another 1week data to test the model. The following data is based on the test data. From the bottom to top, we connect these sensors by Raspberry Pi, Arduino Mega 2560, Arduino Nano; communicate with PC by Serial Port Communication, Bluetooth HC05, WIFI module, cellular network; and store the raw data in “.json” file. We segment the sensor data by the time window (width = 10 s), then obtain the entity event which has been stored into the “.db” file. Alchemy 2.0 is one of the engines to inference by MLN which we have used in this experiment.
Our work is discussing the performance (average precision rate) of the time period for similar activities, like Drink Coffee and Drink Tea. Another work is improving the robustness of the activity model with the missing data and error data. The last work is to compare the Gibbs sampling with MCSAT and find the best method in complex activity situations.
Results and discussion
Similar activities by multiattributes
For example, the terms “definition” and “firstorder logic of DrinkTea and DrinkCoffee” have been shown in the following:
Definition 1
The term of the formula has two classes: variable uses all lowercase letters (e.g., x, y) and constant uses all uppercase letters or number letters (e.g., Bob, 20181225).
Definition 2
The predicate or atom names begin with uppercase letters (e.g., Drink, Work).
Definition 3
The evidence term and query term need to be noted before the inference.
Definition 4
The rules define by the temporalspatial characteristic and the basic traits, the sequence has been realized by like Before predicate.
Description 1 In the firstorder logic, the negation by ¬ symbol, in MLN, is same the ! symbol.
In the experiment, the two activities have more attributes of the time series and time period and the accuracy of similar activities has been improved which is shown in Table 3. Contrasting with the time series and time period rules and traditional rules, the accuracy of similar activities has a significant improvement.
Missing data and disturbance situations
In this part, simplifying the rules by dividing the rules to the common rules and special rules are two parts. Because of the vital role of the special rules, initial weight value is bigger than common rules, the connection weight value between common and special rules is the same with the common rules. We have tested the missing data for different activities, and the accuracy has a significant improvement which has been shown in Table 4.
For example, the simplifying rules of DrinkTea and DrinkCoffee have been shown in the following:
Comparison of Gibbs sampling with MCSAT algorithms
In order to infer the similar activities’ accuracy, we have contrasted the two methods to find the applicable one. In the similar triggered activities’ situation, the traditional MCMC–Gibbs sampling has a good performance in recognizing the two activities simultaneously. The comparison of Gibbs sampling and MCSAT of the two similar interleaved activities has been shown in Table 5. Table 6 is the result of the missing and disturbance situations’ with Gibbs sampling and MCSAT.
On the other hand, improving the accuracy of the missing data or disturbance situations, Gibbs has a better performance than MCSAT. Therefore, for the complex activities, especially for similar activities and missing data or disturbance situations, the Gibbs sampling algorithm has a better performance. The Gibbs sampling algorithm has strong robustness which can be adopted in the future.
Discussion
We can find out from the above experiments that, as the two vital features of an activity, time series and time periods express different time dimensions by operations for the triggered time of entity events.
We can easily find the simplifying rules have extremely good performance of inference by the missing data of one activity, while it has the fluctuations in performance and even has the degradation in the disturbing situation. By comprehensive consideration, simplifying rules have better performance than the original rules.
We can find the Gibbs sampling algorithm has a better performance than MCSAT. Because the two similar activities have many similar rules from entity events, always adopting the MCSAT to handle the independence situations means the two interleaved activities are independent. That generates the conflicting results for the complex activities.
The better result relies on the rules and MLN features. Using the time, time series, periods, and location attributes to represent activity and the simplifying rules, the complex similar and data missing or disturbance situations’ activities can be easily recognized. MLN is the typical hybrid method of the datadriven and knowledgedriven which draws the undirected graph structure that will make all rules more clearly and make the results more accurately.
Conclusion
In this paper, the main work is realizing the complex activity recognition, including the similar activities, data missing, or disturbance situations by Markov logic network. In another important work, we have contrasted the typical methods, Gibbs sampling and MCSAT algorithm, finding the best option in the complex situations. First of all, deploying the multimodeling sensor in our daily life can collect many and varied data to improve the inference accuracy. For data processing, the hierarchical methodology is more efficient for MLN. The structure reduces many repetitive works. Especially for the computational of inference iteration, the exponential grow of the inference prefers the small size of data. In addition, MLN can establish the rule formulas for daily activity based on the time, time series, location, and period features of entity events. The simplifying rules have been adopted to improve the robustness and handle the uncertain situations. Then, MLN is the combination of the firstorder logic and probability, construct Markov network, which has been given weight by the determinative learning method, including the Gibbs sampling and MCSAT algorithms. Afterall, the complex activities have been recognized. We can find that the method has a typical advantage of the soften rules which means even though the personalization habit and data are missing all can be accepted and be solved.
This method is a typical hybrid inference which has the obvious improvement of the complex activities. For inhabitants, the MLN can reflect the preference and habits, except inferencing the activities; mining the relation between these activities is another important work. Also, the next work can go ahead to design the user portrait, especially for the multiresident scenes. From another angle, recognizing the executor of one activity is a point which has a research value. We can do many extend work which will start and prepare based on this work.
Recognition of the daily activity is the fundamental work of many smart devices which is the basics and support for better service for human beings. We will continue to do the related work.
Availability of data and materials
Not applicable.
Abbreviations
 ADL:

Activity of daily living
 DNF:

Disjunctive normal form
 FOL:

Firstorder logic
 MLN:

Markov logic network
 NP:

Nondeterministic polynomial
References
 1
S. Dhelim, H. Ning, M. A. Bouras, J. Ma, in 2018 IEEE SmartWorld, Ubiquitous Intelligence Computing, Advanced Trusted Computing, Scalable Computing Communications, Cloud Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI). Cyberenabled humancentric smart home architecture, (2018), pp. 1880–1886. https://doi.org/10.1109/SmartWorld.2018.00316.
 2
E. Ruiz, V. Osmani, L. E. Sucar, O. Mayora, Detecting dressing failures using temporal–relational visual grammars. J. Ambient. Intell. Humanized Comput. 10(7), 2757–2770 (2019). https://doi.org/10.1007/s1265201809750.
 3
L Lu, H Di, Y Lu, L Zhang, S Wang, Spatiotemporal attention mechanisms based model for collective activity recognition. Signal. ProcessImage Commun. 74:, 162–174 (2019). https://doi.org/10.1016/j.image.2019.02.01.
 4
A. Saeed, T. Ozcelebi, J. Lukkien, Synthesizing and reconstructing missing sensory modalities in behavioral context recognition. Sensors. 18(9) (2018). https://doi.org/10.3390/s1809296.
 5
R. G Hussain, M. A Ghazanfar, M. A Azam, U Naeem, S. U Rehman, A performance comparison of machine learning classification approaches for robust activity of daily living recognition. Artif. Intell. Rev. 52(1), 357–379 (2019). https://doi.org/10.1007/s104620189623.
 6
D Triboan, L Chen, F Chen, Z Wang, A semanticsbased approach to sensor data segmentation in realtime activity recognition. Futur. Gener. Comput. Syst.93:, 224–236 (2019). https://doi.org/10.1016/j.future.2018.09.055.
 7
J. Ye, G. Stevenson, S. Dobson, KCAR: a knowledgedriven approach for concurrent activity recognition. Pervasive Mob. Comput.19:, 47–70 (2015). https://doi.org/10.1016/j.pmcj.2014.02.003.
 8
D. Chakraborty, A. Zaslavsky, in Dependable, Autonomic and Secure Computing, IEEE International Symposium on(DASC), vol. 00. Saguna: Recognizing concurrent and interleaved activities in social interactions, (2011), pp. 230–237. https://doi.org/10.1109/DASC.2011.58. https://doi.ieeecomputersociety.org/10.1109/DASC.2011.58.
 9
D. Tian, X. Xu, Y. Tao, X. Wang, in 2017 IEEE International Conference on Computational Science and Engineering (CSE) and IEEE International Conference on Embedded and Ubiquitous Computing (EUC), vol. 1. An improved activity recognition method based on smart watch data, (2017), pp. 756–759. https://doi.org/10.1109/CSEEUC.2017.148.
 10
M. Babiker, O. O. Khalifa, K. K. Htike, A. Hassan, M. Zaharadeen, in 2017 IEEE 4th International Conference on Smart Instrumentation, Measurement and Application (ICSIMA). Automated daily human activity recognition for video surveillance using neural network, (2017), pp. 1–5. https://doi.org/10.1109/ICSIMA.2017.8312024.
 11
E Kim, S Helal, D Cook, Human activity recognition and pattern discovery. IEEE Pervasive Comput.9(1), 48–53 (2010). https://doi.org/10.1109/MPRV.2010.7.
 12
T. Perumal, Y. L. Chui, M. A. B. Ahmadon, S. Yamaguchi, in 2017 IEEE 6th Global Conference on Consumer Electronics (GCCE). IoT based activity recognition among smart home residents, (2017), pp. 1–2. https://doi.org/10.1109/GCCE.2017.8229478.
 13
Z. Wang, S. Liu, J. Zhang, S. Chen, Q. Guan, A spatio temporal CRF for human interaction understanding. IEEE Trans. Circ. Syst. Video Technol.27(8), 1647–1660 (2017). https://doi.org/10.1109/TCSVT.2016.2539699.
 14
D Sanchez, M Tentori, J Favela, Activity recognition for the smart hospital. IEEE Intell. Syst.23(2), 50–57 (2008). https://doi.org/10.1109/MIS.2008.18.
 15
M. Rodriguez, C. Orrite, C. Medrano, D. Makris, in 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Fast simplexHMM for oneshot learning activity recognition, (2017), pp. 1259–1266. https://doi.org/10.1109/CVPRW.2017.166.
 16
G Okeyo, L Chen, H Wang, Combining ontological and temporal formalisms for composite activity modelling and recognition in smart homes. Futur. Gener. Comput. Syst.39:, 29–43 (2014). https://doi.org/10.1016/j.future.2014.02.014. Special Issue on Ubiquitous Computing and Future Communication Systems.
 17
B. Bouchard, S. Giroux, A. Bouzouane, in Advances in Artificial Intelligence, ed. by L. Lamontagne, M Marchand. A smart home agent for plan recognition (SpringerBerlin, 2006), pp. 25–36.
 18
C. Villalonga, R. M. Asif, A. K. Wajahat, H. Pomares, I. Rojas, S. Lee, B. L. Oresti, Ontologybased highlevel context inference for human behavior identification. Sensors (Switserland). 16(10), 1–26 (2016). https://doi.org/10.3390/s16101617.
 19
L. Chen, C. Nugent, G. Okeyo, An ontologybased hybrid approach to activity modeling for smart homes. IEEE Trans. Hum.Mach. Syst.44(1), 92–105 (2014). https://doi.org/10.1109/THMS.2013.2293714.
 20
G. Civitarese, C. Bettini, T. Sztyler, D. Riboni, H Stuckenschmidt, newnectar: collaborative active learning for knowledgebased probabilistic activity recognition. Pervasive Mob. Comput.56: (2019). https://doi.org/10.1016/j.pmcj.2019.04.006.
 21
H. S. Shin, D. Turchi, S. He, A. Tsourdos, Behavior monitoring using learning techniques and regularexpressionsbased pattern matching. IEEE Trans. Intell. Transp. Syst.20(4), 1289–1302 (2019). https://doi.org/10.1109/TITS.2018.284926.
 22
T. Darpan, C. Liming, C. Feng, W. Zumin, Semantic segmentation of realtime sensor data stream for complex activity recognition. Pers. Ubiquit. Comput.21(3), 411–425 (2017). https://doi.org/10.1007/s0077901710055.
 23
V. Goranko, Logic in computer science: modelling and reasoning about systems. J. Logic Lang. Inf.16(1), 117–120 (2007).
 24
B. Pfahringer, ed. by C. Sammut, G. I Webb. Conjunctive Normal Form (SpringerBoston, 2017), pp. 260–261.
 25
P. Singla, P. Domingos, in Proceedings, The Twentieth National Conference on Artificial Intelligence and the Seventeenth Innovative Applications of Artificial Intelligence Conference. Discriminative Training of Markov Logic Networks[C] (AAAI PressPittsburgh, 2005), pp. 868–873.
 26
D. Lowd, P. Domingos, in Knowledge discovery in databases: PKDD 2007, ed. by J. N. Kok, J. Koronacki, R. Lopez de Mantaras, S. Matwin, D. Mladeniċ, and A. Skowron. Efficient weight learning for markov logic networks (SpringerBerlin, 2007), pp. 200–211.
 27
P. Domingos, S. Kok, D. Lowd, H. Poon, M. Richardson, P. Singla, ed. by L. De Raedt, P. Frasconi, K. Kersting, and S Muggleton. Markov Logic (SpringerBerlin, 2008), pp. 92–117.
 28
L. Getoor, B. Taskar, Markov Logic: a Unifying Framework for Statistical Relational Learning, vol. 0 (MITP, Cambridge, 2007).
 29
S. D. Tran, L. S. Davis, in Computer Vision – ECCV 2008, ed. by D. Forsyth, P. Torr, and A. Zisserman. Event modeling and recognition using markov logic networks (SpringerBerlin, 2008), pp. 610–623.
 30
T. Geier, S. Biundo, in 2011 IEEE 23rd International Conference on Tools with Artificial Intelligence. Approximate online inference for dynamic Markov logic networks, (2011), pp. 764–768. https://doi.org/10.1109/ICTAI.2011.120.
 31
H. Ning, F. Shi, T. Zhu, Q. Li, L. Chen, A novel ontology consistent with acknowledged standards in smart homes. Comput. Netw.148:, 101–107 (2019). https://doi.org/10.1016/j.comnet.2018.11.004.
 32
D. Riboni, T. Sztyler, G. Civitarese, H. Stuckenschmidt, in Proceedings of the 2016 ACM International Joint Conference on Pervasive and Ubiquitous Computing. UbiComp ’16. Unsupervised recognition of interleaved activities of daily living through ontological and probabilistic reasoning (ACMNew York, 2016), pp. 1–12. https://doi.org/10.1145/2971648.2971691.
 33
K. S. Gayathri, S. Elias, R. Balaraman, Hierarchical activity recognition for dementia care using Markov logic network. Personal. Ubiquit. Comput.19(2), 271–285 (2015). https://doi.org/10.1007/s0077901408277.
Acknowledgements
Not applicable.
Funding
This work was supported by the National Natural Science Foundation of China (61872038 and 61811530335), and in part by the Fundamental Research Funds for the Central Universities under Grant FRFBD18016A, and UK Royal SocietyNewton Mobility Grant (No.IEC\NSFC\170067).
Author information
Affiliations
Contributions
QL designed the solution, wrote this paper, and did the experimental tests. SC assisted the analysis of the algorithm. HN, TZ, and LC guided and checked the whole paper, experiments, and figures. All authors read and approved the final manuscript.
Corresponding author
Correspondence to Huansheng Ning.
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License(http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Cite this article
Li, Q., Ning, H., Zhu, T. et al. A hybrid approach to inferring the Internet of Things for complex activity recognition. J Wireless Com Network 2019, 251 (2019) doi:10.1186/s1363801915537
Received:
Accepted:
Published:
Keywords
 Sensors
 Complex activity recognition
 Markov logic network
 Data missing or disturbance