Unmanned aerial vehicle (UAV)-enabled mobile edge computing (MEC) networks encounter significant challenges in achieving balanced workload distribution, primarily due to the limited coverage areas of UAVs and their diverse computational capabilities. This paper proposes a UAV-enabled MEC framework that jointly optimizes three-dimensional (3D) trajectory planning and dynamic computation offloading. We formulate a mixed-integer programming (MIP) problem to minimize system latency by simultaneously optimizing UAV trajectory design and task offloading strategies, where UAV mobility and offloading decisions are tightly coupled.Unlike existing approaches that either optimize 3D trajectories without inter-UAV cooperation or implement cooperative computing under fixed altitudes with predetermined relay hops, our framework uniquely integrates adaptive multi-hop collaborative offloading with continuous 3D trajectory planning. The model complexity arises from its hybrid decision structure that simultaneously handles continuous trajectory parameters and discrete offloading variables. Our approach decomposes the problem into two tightly coupled subproblems: (1) 3D UAV trajectory optimization and (2) task offloading scheduling. We then propose a Decoupled Deep Reinforcement Learning for Parallelized Planning and Offloading (DDP3O) algorithm that systematically addresses these interconnected components. Experimental results demonstrate that DDP3O achieves fast convergence and superior performance compared to state-of-the-art methods including block coordinate descent optimization, DQN-based approaches, and fixed-hop cooperative schemes across multiple operational scenarios.

Recent advances in Large Language Models (LLMs) have revolutionized mobile robots, including unmanned aerial vehicles (UAVs), enabling their intelligent operation within Internet of Things (IoT) ecosystems. However, LLMs still face challenges from logical reasoning and complex decision-making, leading to concerns about the reliability of LLM-driven UAV operations in IoT applications. In this paper, we propose a closed-loop LLM-driven UAV operation code generation framework that enables reliable UAV operations powered by effective feedback and refinement using two LLM modules, i.e., a Code Generator and an Evaluator. Our framework transforms numerical state observations from UAV operations into semantic trajectory descriptions to enhance the evaluator LLM’s understanding of UAV dynamics for precise feedback generation. Our framework also enables a simulation-based refinement process, and hence eliminates the risks to physical UAVs caused by incorrect code execution during the refinement. Extensive experiments on UAV control tasks with different complexities are conducted. The experimental results show that our framework can achieve reliable UAV operations using LLMs, which significantly outperforms baseline methods in terms of success rate and completeness with the increase of task complexity.

Retrieval-Augmented Generation (RAG) empowers Large Language Models (LLMs) to dynamically integrate external knowledge during inference, improving their factual accuracy and adaptability. However, adversaries can inject poisoned external knowledge to override the model’s internal memory. While existing attacks iteratively manipulate retrieval content or prompt structure of RAG, they largely ignore the model’s internal representation dynamics and neuron-level sensitivities. The underlying mechanism of RAG poisoning has not been fully studied and the effect of knowledge conflict with strong parametric knowledge in RAG is not considered. In this work, we propose NeuroGenPoisoning, a novel attack framework that generates adversarial external knowledge in RAG guided by LLM internal neuron attribution and genetic optimization. Our method first identifies a set of **Poison-Responsive Neurons** whose activation strongly correlates with contextual poisoning knowledge. We then employ a genetic algorithm to evolve adversarial passages that maximally activate these neurons. Crucially, our framework enables massive-scale generation of effective poisoned RAG knowledge by identifying and reusing promising but initially unsuccessful external knowledge variants via observed attribution signals. At the same time, Poison-Responsive Neurons guided poisoning can effectively resolves knowledge conflict. Experimental results across models and datasets demonstrate consistently achieving high Population Overwrite Success Rate (POSR) of over 90\% while preserving fluency. Empirical evidence shows that our method effectively resolves knowledge conflict.

Reconfigurable Intelligent Surfaces (RIS) can optimize spectrum and energy efficiency in the sixth-generation (6G) wireless communication system through dynamic electromagnetic wave manipulation. The programmable control of spatial electromagnetic signals by RIS presents a double-edged sword, and it can also be exploited by malicious attackers. However, few studies have focused on the detection and identification of such malicious RIS. To fill this gap, we propose a novel spoofing detection framework combining dynamic key-embedded phase codebooks with a dual-channel feature extraction mechanism. This approach jointly decodes wireless channel fingerprints and cryptographic signatures from received signals. A hybrid discriminator, integrating autoencoder-based signal reconstruction fidelity and key-matching validation, enables robust legitimacy verification. The prototype experiments using USRP SDR and RIS hardware show that the verification accuracy of the scheme can reach 100%, when the signal-to-noise ratio (SNR) is above 10dB, the number of training sample points is more than 128, and the codebook dimension is near 32.

Wireless device pairing is a critical security mechanism to bootstrap the secure communication between two devices without a pre-shared secret. It has been widely used in many Internet of Things (IoT) applications, such as smarthome and smarthealth. Most existing device pairing mechanisms are based on out-of-band channels, e.g., extra sensors or hardware, to validate the location proximity of pairing devices. However, out-of-band channels are not universal on all wireless devices, thus this type of scheme is limited to certain application scenarios or conditions. On the other hand, in-band channel-based device pairing aims at universal applicability by only relying on wireless interfaces. Existing in-band channel-based pairing schemes either require multiple antennas separated in a good distance on one pairing devices which is not applicable in certain scenarios, or require users to repeat multiple sweeps which is not optimal in terms of usability. Therefore, an in-band wireless device pairing scheme providing high security while maintaining good usability (simple pairing process and user interaction) is highly desired. In this work, we propose an easy-to-use mutual authentication device pairing scheme, named Swipe2Pair, based on location proximity of pairing devices and wireless transmission power randomization. We conduct extensive security analysis and collect considerable experimental data under various settings in different environments. Experimental results show that Swipe2Pair achieves high security and usability. It only takes less than one second to complete the pairing process with a simple swipe of one device in front of the other.

Federated learning (FL) enables multiple clients to jointly train a global learning model while keeping their training data locally, thereby protecting clients’ privacy. However, there still exist some security issues in FL, e.g., the honest-but-curious servers may mine privacy from clients’ model updates, and the malicious clients may launch poisoning attacks to disturb or break global model training. Moreover, most previous works focus on the security issues of FL in the presence of only honest-but-curious servers or only malicious clients. In this paper, we consider a stronger and more practical threat model in FL, where the honest-but-curious servers and malicious clients coexist, named as the non-fully trusted model. In the non-fully trusted FL, privacy protection schemes for honest-but-curious servers are executed to ensure that all model updates are indistinguishable, which makes malicious model updates difficult to detect. Toward this end, we present an Adaptive Privacy-Preserving FL (Ada-PPFL) scheme with Differential Privacy (DP) as the underlying technology, to simultaneously protect clients’ privacy and eliminate the adverse effects of malicious clients on model training. Specifically, we propose an adaptive DP strategy to achieve strong client-level privacy protection while minimizing the impact on the prediction accuracy of the global model. In addition, we introduce DPAD, an algorithm specifically designed to precisely detect malicious model updates, even in cases where the updates are protected by DP measures. Finally, the theoretical analysis and experimental results further illustrate that the proposed Ada-PPFL enables client-level privacy protection with 35% DP-noise savings, and maintains similar prediction accuracy to models without malicious attacks.

Ambient backscatter communication (AmBC) enables ultra-low-power communications by backscattering ambient radio frequency (RF) signals and harvesting energy simultaneously. It has emerged as a cutting-edge technology for supporting a variety of Internet of Things (IoT) applications. However, existing research lacks effective secret key sharing schemes for safeguarding communications between resource-constrained backscatter devices (BDs) in AmBC systems. In this paper, we present, Tri-Channel, a novel physical layer key generation scheme between two BDs by multiplying downlink signals and backscatter signals to obtain the information of a triangle channel as a shared random secret source for key generation. In particular, we analyze the security of our scheme under both passive and active attacks, concretely Eavesdropping Attack (EA), Control Channel Attack (CCA), Signal Manipulative Attack (SMA), and Untrusted RF-Source Attack (URSA). Through theoretical analysis and simulations by comparing with a traditional scheme (named Tradi-Channel), we found that our scheme consistently outperforms the Tradi-Channel under the EA and two active attacks (CCA and SMA). In addition, it shows better security performance under URSA, which is proposed based on the unauthenticated characteristic of BDs in Tri-Channel, even though URSA is more vital than SMA. Concretely, Tri-Channel’s secret key rate (SKR) outperforms Tradi-Channel’s under the above four passive and active attacks. This implies that our scheme is advanced in terms of both security and efficiency of key generation. Numerous extensive simulations further prove our theoretical analysis results.

With the increasing demands for sharing confidential information among massive Internet of Things (IoT) devices in 5G and beyond wireless networks, many applications require the common secret key generation for a group of IoT devices. However, most of the existing works on physical layer secret key generation (PLKG) only focus on the pairwise key generation between two users, which is a low efficient and high cost to be extended to the scenarios of group key generation. In this work, we propose a new efficient multiple-input–multiple-output (MIMO) physical layer group secret key generation scheme to reduce the consumption of channel probing and improve the efficiency for group key generation. Different from current schemes, in the proposed scheme, the transmitter randomly generates the group secret key and directly broadcasts the downlink data symbols to the group users. At the receiver end, each group user can efficiently “observe” the common group key through the downlink broadcasting data symbols, while keeping perfect secrecy of the shared group key against eavesdroppers. The performance of reliability, security, and the group key generation rate is fully investigated, which shows the advantages of high efficiency, low consumption, and strong robustness of the proposed scheme. Extensive simulations are conducted to validate the effectiveness of the proposed scheme.

Physical-layer fingerprinting is a promising technique to identify Internet of Things (IoT) devices. In this article, we investigate a new radio-frequency (RF) fingerprinting based on the distinctive signal-to-noise-ratio (SNR) trace in the sector-level sweep (SLS) procedure of 5G IEEE 802.11ad devices. This SLS SNR trace-based fingerprinting can directly apply to off-the-shelf devices without any extra hardware requirements and be independent of the wireless channel and environment. To tackle the impact of orientation on the RF fingerprinting, we propose a novel fingerprinting framework, involving correlation analysis, surface fitting, curve pursuing, and binary classification, named the CSCB framework. Using this framework, the proposed SLS SNR trace-based fingerprinting can achieve device authentication at any orientation with one receiver under line-of-sight (LOS) or non-LOS (NLOS) scenarios. We conduct proof-of-concept experiments using off-the-shelf IEEE 802.11ad devices (Talon AD7200 and MG360 WiGig) to evaluate the performance of the proposed fingerprinting schemes. Experimental results show the effectiveness of the proposed fingerprinting schemes where the verification accuracy of the proposed scheme can reach 99% with only 200 training samples.

Identity-based attacks (IBAs) are one of the most serious threats to wireless networks. Recently, there is an increasing interest in using the received signal strength (RSS) to detect IBAs in wireless networks. However, current schemes tend to generate excessive false alarms in the mobile scenario. In this paper, we propose a stronger Reciprocal Channel Variation-based Identification and classification (RCVIC) scheme for the mobile wireless networks, which exploits the reciprocity of the wireless fading channel and RSS variations naturally incurred by mobility to improve the detection performance. Different from current schemes only detect IBAs, RCVIC scheme conducts a multi-stage detection processes. If the IBAs are detected, RCVIC scheme partitions the received frames into two classes. The frames in the same class should be sent from the same senders, which could benefit the further analysis, such as network forensics, attacker localizing and trajectory analysis, etc. The feasibility of RCVIC are numerically evaluated through theoretical analysis and simulations. It is further validated through experiments using off-the-shelf 802.11 devices under different attacking patterns in real indoor and outdoor mobile scenarios.

Recently, networks operate at frequencies over 28 GHz (mmWave) have emerged as a viable solution for 5G mobile networks to provide Gbps data rate. Due to the high directivity and attenuation of mmWave signals, mmWave communication links are highly vulnerable to the frequent mmWave channel blockages, which can trigger excessive handovers. Thanks to its ability to enrich the scattering environment and create reflective signal multipaths, Reconfigurable Intelligent Surface (RIS) has great potential to counter the blockage effect and thus greatly reduce the number of unnecessary handovers. However, this potential has not been well explored. In this paper, we propose a RIS-assisted handover scheme by leveraging deep reinforcement learning (DRL). Under various channel blockage conditions, the DRL agent manages to reduce the cumulative handover overhead by jointly adjusting beamformers and RIS phase shifts. Compared with the existing schemes without considering RIS, the RIS-assisted handover scheme significantly reduces the number of handovers and achieves higher spectrum efficiency. Besides, to alleviate the impact from the limited observations of the fast fading channels, we propose a lightweight algorithm to sense the blockage status and such sensing results can be utilized to improve the performance of model training. Numerical results show that DRL agent is able to further improve the performance when integrated with the blockage status sensing algorithm.

Ambient backscatter communication (AmBC), which enables energy harvesting and ultra-low-power communication by utilizing ambient radio frequency (RF) signals, has emerged as a cutting-edge technology to realize numerous Internet of Things (IoT) applications. However, the current literature lacks efficient secret key sharing solutions for resource-limited devices in AmBC systems to protect the backscatter communications, especially for private data transmission. Thus, we propose a novel physical layer key generation scheme between backscatter devices (BDs) by exploiting received superposed ambient signals. Based on the repeated patterns (i.e., cyclic prefix in OFDM symbols) in ambient RF signals, we present a joint transceiver design of BD backscatter waveform and BD receiver to extract the downlink signal and the backscatter signal from the superposed signals. By multiplying the downlink signal and the backscatter signal, we can actually obtain the triangle channel information as a shared random secret source for key generation. Besides, we study the trade-off between the rate of secret key generation and harvested energy by modeling it as a joint optimization problem. Finally, extensive numerical simulations are provided to evaluate the key generation performance, energy harvesting performance, and their trade-offs under various system settings.

We study security of remote state estimation in wireless cyber-physical systems (CPS) where a sensor sends its measurements to the remote state estimator over a multichannel wireless link in presence of a jamming attacker. Most of the existing works study the sensor's defense scheme by adopting optimization-based methods and rely on the prior knowledge of the attacker's attack policy. To relax this constraint, we propose a novel online-learning-based policy called joint channel and power selection (J-CAP) for the sensor to dynamically choose transmission channel and power. The proposed method assumes no prior knowledge of the attacker's attack policy, nor of the channel state information. J-CAP jointly optimizes sensor's channel selection and power consumption, and guarantees the estimator's asymptotic stability. We theoretically prove that J-CAP achieves a sublinear learning regret bound. We also show J-CAP's optimality by deriving and matching its regret lower and upper bound orders. Compared with the solution that directly applies the baseline solution, J-CAP improves the regret upper bound by a factor of √{K+L}, where K and L denote the number of channels and number of power levels, respectively. Numerical evaluations validate the analytical results under various CPS parameters, and compare the J-CAP's performance with the state-of-the-art solutions.

Recently, MIMO (multiple-input-multiple-output) physical layer secure transmission has attracted great attentions. However, current schemes cannot defend against the passive eavesdroppers with arbitrary number of antennas. To address this problem, in this work, we propose a practical physical layer MIMO secure communication scheme (PLSC) to defend against such an eavesdropper with arbitrary number of antennas. In the proposed scheme, the transmitter first independently generates a random binary sequence as the “key bits (KB)” to “encrypt” (XOR) the confidential information. After that, the transmitter sends the “encrypted information” over the wireless channel, along with mapping key bits to the legitimate receiver simultaneously. The key principle lies in that the KB information is coded in the indexes of the activated/non-activated antennas combination of the legitimate user. Then, the legitimate receiver first observes his/her activated antenna indexes to obtain the corresponding key bits. After that, he/she demodulates the “encrypted information” at the activated antennas, and finally “decrypts” (XOR) the confidential information by using the observed key bits. However, due to the uniqueness and independence of MIMO wireless channel, for any other eavesdroppers who suffer an independent channel from legitimate users, we prove that it cannot observe any information about KB from the received signals, regardless of how many antennas it has used. Consequently, without knowledge of KB, it cannot decrypt any information about the confidential information, too. The reliability and security of PLSC are theoretically demonstrated. The simulation and numerical results fully verified the validity and effectiveness of the proposed scheme.

Grant-free random access is an emerging technology for providing massive connectivity for 5G massive machine-type communications (mMTC), where non-orthogonal pilot sequences are used to simultaneously detect active users and estimate channels. However, grant-free 5G IoT networks are vulnerable to pilot contamination attacks (PCA), where the attacker can send the same pilots as legitimate IoT users to harm the active user detection and channel estimation. To defend against this attack, in this article, we propose a physical-layer countermeasure based on the channel virtual representation (CVR). CVR can emphasize the unique characteristics of mmWave channels that are sensitive to the location of the sender. This can be utilized to counter PCA no matter if the attacker's pilots are superimposed to that of the victim or not. Based on this observation, to achieve an efficient PCA detection, a single-hidden-layer multiple measurement (SHMM) Siamese network is employed. This solution tackles the challenges of channel randomness and massive connectivity in mMTC IoT networks, and supports small sample learning. Simulation results evaluate and confirm the effectiveness of the proposed detection scheme under various scenarios. The detection accuracy can approach 99% with 128 antennas at the receiver and reach above 95% even with only 50 training samples.

Spoofing attacks pose a serious threat to wireless communications. Exploiting physical-layer features to counter spoofing attacks is a promising solution. Although various physical-layer spoofing attack detection (PL-SAD) techniques have been proposed for conventional 802.11 networks in the sub-6GHz band, the study of PL-SAD for 802.11ad networks in 5G millimeter wave (mmWave) 60GHz band is largely open. In this paper, to achieve efficient PL-SAD in 5G networks, we propose a unique physical layer feature in IEEE 802.11ad networks, i.e., the signal-to-noise-ratio (SNR) trace obtained at the receiver in the sector level sweep (SLS) process. The SNR trace is readily extractable from the off-the-shelf device, and it is dependent on both transmitter location and intrinsic hardware impairment. Therefore, it can be used to achieve an efficient detection no matter the attacker is co-located with the legitimate transmitter or not. To achieve spoofing attack detection, we provide two methods based on different machine learning models. For the first method, the detection problem is formulated as a machine learning classification problem. To tackle the small sample learning and fast model construction challenges, we propose a novel neural network framework consisting of a backpropation network, a forward propagation network, and generative adversarial networks (GANs). Another method involves a Siamese network, in which the similarity between sample pairs from one device is used to achieve PL-SAD. It can tackle the training problem that the historical data cannot support the identification of the same device in a new communication session. We conduct experiments using off-the-shelf 802.11ad devices, Talon AD7200s and MG360, to evaluate the performance of the proposed PL-SAD schemes. Experimental results confirm the effectiveness of the proposed PL-SAD schemes, and the detection accuracy can reach 99% using small sample sizes under different scenarios.

Spoofing attacks pose a serious threat to wireless communications. Exploiting physical-layer features to counter spoofing attacks is a promising solution. Although various physical-layer spoofing attack detection (PL-SAD) techniques have been proposed for conventional 802.11 networks in the sub-6GHz band, the study of PL-SAD for 802.11ad networks in 5G millimeter wave (mmWave) 60GHz band is largely open. In this paper, we propose a unique physical layer feature in IEEE 802.11ad networks, i.e., the signal-to-noise-ratio (SNR) trace obtained at the receiver in the sector level sweep (SLS) process, to achieve efficient PL-SAD. The SNR trace is readily extractable from the off-the-shelf device, and it is dependent on both transmitter location and intrinsic hardware impairment. Therefore, it can be used to achieve an efficient detection no matter the attacker is co-located with the legitimate transmitter or not. The detection problem is formulated as a machine learning classification problem. To tackle the small sample learning and fast model construction challenges, we propose a novel neural network framework consisting of a backpropation network, a forward propagation network, and generative adversarial networks (GANs). It can tackle small sample learning and allow for quick model construction. We conduct experiments using off-the-shelf 802.11ad devices, Talon AD7200s and MG360, to evaluate the performance of the proposed PL-SAD scheme. Experimental results confirm the effectiveness of the proposed PL-SAD scheme, and the detection accuracy can reach 98% using small sample sizes under different scenarios.

Power non-orthogonal multiple access (NOMA) has been considered as a new enabling technology in 5G communication. In this paper, we introduce the problem of pilot contamination attack (PCA) on NOMA in millimeter wave (mmWave) and massive MIMO 5G communication. Due to the new characteristics of NOMA such as superposed signals with multi-users, PCA detection faces new challenges. By harnessing the sparseness and statistics of mmWave and massive MIMO virtual channel, we propose two effective PCA detection schemes for NOMA tackling static and dynamic environments, respectively. For the static environment, the problem of PCA detection is formulated as a binary hypothesis test of the virtual channel sparsity. For the dynamic environment, the statistic of the peaks in the virtual channel is leveraged to distinguish the contamination state from the normal state. A peak estimation algorithm and a machine learning based detection framework are proposed to achieve high detection performance. To further optimize the proposed scheme, a feature selection algorithm and an optimization model considering the detection accuracy and detection delay are presented. Simulation results evaluate and confirm the effectiveness of the proposed detection schemes. The detection rate can approach 100% with 10-3false alarm rate in the static environment and above 95% in the dynamic environment under various system parameters.

The fifth generation (5G) wireless technologies serve as a key propellent to meet the increasing demands of the future Internet of Things (IoT) networks. For wireless communication security in 5G IoT networks, physical-layer security (PLS) has recently received growing interest. This paper aims to provide a comprehensive survey of the PLS techniques in 5G IoT communication systems. The investigation consists of four hierarchical parts. In the first part, we review the characteristics of 5G IoT under typical application scenarios. We then introduce the security threats from the 5G IoT physical-layer and categorize them according to the different purposes of the attacker. In the third part, we examine the 5G communication technologies in 5G IoT systems and discuss their challenges and opportunities when coping with physical-layer threats, including massive multiple-input-multiple-output (MIMO), millimeter wave (mmWave) communications, nonorthogonal multiple access (NOMA), full-duplex technology, energy harvesting (EH), visible light communication (VLC), and unmanned aerial vehicle (UAV) communications. Finally, we discuss open research problems and future works about PLS in the IoT system with technologies of 5G and beyond.

In this work, we present a scheme of physical layer secret key generation for Millimeter wave (mmWave) Massive MIMO system. Our scheme is compatible with current hardware structure and protocols including Analog Beamforming, Massive MIMO, and Beam Sweep. We add a small perturbation angle into the Angle of Arrival (AoA) of the transmitter as the common randomness, which significantly improved the secret key rate without being constrained by the complexity of link initialization protocols and low dynamic of the channel. Therefore, the secret key rate can be enhanced by increasing the number of perturbations. In addition, our scheme can combat co-located eavesdropper (Eve) by utilizing the high directionality of Massive MIMO antenna. Numerical results show that our scheme has a high bit agreement ratio (BAR) between legitimate users while the co-located Eve only gets the BAR around 50%, which indicates that the secrecy of the generated key is well achieved.

Physical layer security of non-orthogonal multiple access (NOMA) systems has attracted great attentions. However, the impact of mobility on physical layer security of NOMA systems has not been well studied. In this paper, to fill this gap, we investigate the impact of random mobility on physical layer security of NOMA systems. Considering scenarios where a base station (BS) or access point (AP) communicates to two random mobile users with a passive eavesdropper in two concentric circles, we study the secrecy performance with combinations of two typical random mobility models: random waypoint (RWP) and random direction (RD). A general analytical framework to numerically calculate the average secrecy rates of NOMA mobile users under steady state is provided. By comparing secrecy performance of mobile users with static users, we find that RWP mobile users can achieve higher average secrecy rates than the users with other mobility combinations. Meanwhile, two types of secrecy fairness for mobile users are fully considered and we propose a novel sum average secrecy rate maximization problem, subject to average power limits and users' QoS (quality of service) requirements. Considering eavesdropper's channel state information (CSI) is unknown to BS, we propose a threshold power allocation strategy to improve the sum average secrecy rate of NOMA mobile users. Extensive numerical simulations are conducted to validate our model and theoretical analysis.

In many IoT (Internet-of-Things) applications, a large number of low-cost IoT devices are connected to the Internet through an access point (AP) or gateway via wireless communication. Due to the resource constraints on IoT devices and broadcast nature of wireless medium, identity spoofing attacks are easy to launch but hard to defend in an IoT wireless access network. In this paper, under the context of 5G communication, we propose an efficient physical layer identity spoofing attack detection scheme for IoT. By harnessing the sparsity of the virtual channel in mmWave and Massive MIMO 5G communication, we propose a two- step detection scheme. In the first step, our scheme detects anomalies by examining the virtual angles of arrival (AoA) and path gains of all the IoT devices simultaneously in a virtual channel space (VCS). In the second step, we introduce a machine learning based detection scheme to detect the actual attack. Simulation results evaluate and confirm the effectiveness of the proposed detection scheme. The minimum Bayes risk of the proposed scheme can be less than 0.5\% even in the presence of 100 IoT devices.

Using cameras for detecting hazardous or suspicious events has spurred new research for security concerns. To make such detection reliable, researchers trust overcome difficulties such as variation in camera capabilities, environmental factors. imbalances of positive and negative training data, and asymmetric costs of misclassifying events of different classes. Following up on the event-detection framework (Wu et al. (2003)) that we have proposed, we present in this paper the framework's two major components: invariant feature extraction and biased statistical inference. We report results of our experiments using the framework for detecting suspicious motion events in a parking lot.

We present a framework for multi-camera video surveillance. The framework consists of three phases: detection, representation, and recognition. The detection phase handles multi-source spatio-temporal data fusion for efficiently and reliably extracting motion trajectories from video. The representation phase summarizes raw trajectory data to construct hierarchical, invariant, and content-rich descriptions of the motion events. Finally, the recognition phase deals with event classification and identification on the data descriptors. Because of space limits, we describe only briefly how we detect and represent events, but we provide in-depth treatment on the third phase: event recognition. For effective recognition, we devise a sequence-alignment kernel function to perform sequence data learning for identifying suspicious events. We show that when the positive training instances (i.e., suspicious events) are significantly outnumbered by the negative training instances (benign events), then SVMs (or any other learning methods) can suffer a high incidence of errors. To remedy this problem, we propose the kernel boundary alignment (KBA) algorithm to work with the sequence-alignment kernel. Through empirical study in a parking-lot surveillance setting, we show that our spatio-temporal fusion scheme and biased sequence-data learning method are highly effective in identifying suspicious events.