The emergence of Vision-Language Models (VLMs) like Contrastive Language-Image Pretraining (CLIP) provides appealing solutions to various vision problems including Dynamic Facial Expression Recognition (DFER). However, most of the proposed approaches face major challenges, particularly related to inefficient full fine-tuning of the encoders and the complexity of the models. Moreover, some of the proposed methods seem to struggle with suboptimal performance due to (i) poor alignment between textual and visual representations, and (ii) ineffective temporal modeling. To address these challenges, we propose PE-CLIP, a parameter-efficient fine-tuning (PEFT) framework that elegantly adapts CLIP for dynamic facial expression recognition, requiring significantly reduced number of trainable parameters while maintaining high accuracy. At its core, to enhance efficiency and performance, PE-CLIP introduces two specialized adapters namely a Temporal Dynamic Adapter (TDA) and a Shared Adapter (ShA). The TDA is a GRU-based module with a dynamic scaling mechanism, capturing sequential dependencies while adaptively modulating the contribution of each temporal feature to emphasize the most informative ones while mitigating irrelevant variations. The ShA is a lightweight adapter refine representations within both textual and visual encoders, ensuring consistent feature processing while maintaining parameter efficiency. Additionally, we leverage Multi-modal Prompt Learning (MaPLe), which introduces learnable prompts to both visual and action unit-based textual description inputs, further improving the semantic alignment between modalities and enabling the efficient adaptation of CLIP for dynamic tasks. We evaluate our proposed PE-CLIP on two benchmark datasets, namely DFEW, FERV39K, and AFEW, achieving competitive performance compared to state-of-the-art methods while requiring fewer trainable parameters. By striking an optimal balance between parameter efficiency and performance, PE-CLIP sets a new benchmark in resource-efficient DFER. The source code of the proposed PE-CLIP will be publicly available at https://github.com/Ibtissam-SAADI/PE-CLIP .

This paper introduces a novel hybrid approach for noise reduction in B-scan UWB radar data, specifically designed for the detection of cyclists, pedestrians, and cars. The proposed method integrates Higher-Order Statistics (HOS) with Wavelet Shrinkage Denoising (WSD) to enhance signal quality and improve detection accuracy. The effectiveness of this hybrid approach is evaluated against widely used noise removal methods, including Principal Component Analysis (PCA), Singular Value Decomposition (SVD), and the individual use of WSD and HOS. A quantitative analysis based on the Signal-to-Noise Ratio (SNR) demonstrates that the HOS-WSD combination outperforms the other methods in terms of noise reduction. These results highlight the superiority of the hybrid approach and its potential to enhance the robustness of UWB radar detection systems.

International audience

Accurate environmental perception is crucial for the safe and efficient navigation of autonomous vehicles. Integrating data from various sensors, such as cameras, radars, and lidars, presents challenges due to differing sensor reliability and environmental conditions. This study introduces a new method for combining data from multiple sensors using evidence theory (Dempster-Shafer), which helps to address uncertainties and conflicting information. By dynamically merging sensor inputs and resolving conflicts, our model improves object detection and classification in uncertain situations. Through extensive simulations and detailed analyses using metrics like Belief (Bel), Plausibility (Pl), Confidence Interval (CI), and Pignistic Probability (P), the study confirms the effectiveness and dependability of this fusion approach for practical autonomous vehicle applications.

This paper introduces a novel localization framework for GPS-denied environments, leveraging reflective QR codes, multi-sensor fusion, and V2X communication. The system integrates LiDAR-detectable QR codes encrypted with GPS coordinates, dead reckoning, and V2X data sharing to enhance localization accuracy in challenging conditions such as tunnels. A factor graph-based optimization fuses data from IMU, QR code detections, and V2X communication to ensure robustness. Experimental results demonstrate significant localization accuracy improvements over conventional methods, reducing position errors while maintaining reliable performance in lowlight conditions. The proposed approach offers a scalable and secure solution for autonomous vehicle localization, ensuring continuous operation and data integrity through end-to-end encryption.

International audience

ECHO is a next-generation advanced driverassistance system (ADAS) that integrates multi-modal perception, cooperative V2X communication, and an agentic large language model (LLM) for explainable, human-centric support. Unlike traditional ADAS, ECHO does not directly control the vehicle but interprets real-time data from sensors, infrastructure, and driver state, producing transparent natural language guidance and standardized V2X intent messages. Extensive evaluation in both simulation and real-world settings shows that ECHO reduces reaction times to V2X events by over 40 %, decreases near-miss incidents by 60 %, and significantly enhances cooperative maneuver success compared to conventional ADAS. These results demonstrate the potential of agentic, explainable AI to advance connected mobility with improved safety, trust, and driver engagement.

This paper presents a novel localization approach that enhances accuracy and reduces error drift by integrating High Definition (HD) maps with multi-sensor data. Leveraging LiDAR and GPS data, our method cross-references detected physical landmarks with HD maps to improve precision. When sensor data is obstructed, Vehicle-to-Infrastructure (V2I) communication ensures continued accuracy. This hybrid approach significantly improves localization, especially in autonomous driving and urban planning. Our solution sets a new standard for real-time localization, enhancing the safety and efficiency of autonomous vehicle operations.

The development of autonomous vehicles has garnered significant attention in recent years due to its potential to revolutionize the way we move and its impact on society, security, and the environment. One of the crucial components of these vehicles is the object detection system, responsible for identifying and localizing objects in the road environment. This task is essential for decision-making processes in autonomous vehicles, such as navigating the vehicle, avoiding obstacles, and changing the driving direction.Despite the challenges posed by the variability of objects in the road environment, this paper proposes a system for autonomous vehicle guidance based on embedded systems and the RTMAPS tool. The primary focus of the system is to perform effective object detection.To achieve this objective, the study evaluates several stateof-the-art algorithms for 2D object detection and selects the best algorithms based on precision and inference time. The selected algorithm is then implemented using RTMAPS in realworld scenarios on the university’s track. The results of this study provide valuable insights for the research and practical community in the field of autonomous vehicles and serve as a reference for future work in object detection

Several works are aiming to develop techniques allowing detecting False Data Injection Attacks, which represents one of the most harmful attacks due to its ability to damage a Cyber Physical Systems (CPS). Among these techniques the watermarking represents one of the most used ones. This paper proposes the design of a False Data Injection Attack (FDIA) against a CPS protected by a watermark-based detector. The attack herein proposed is achieved in two phases. The first one is a passive phase, where the adversary builds a black box model of the system. Then, he uses the already built model to create the FDIA without being detected by the watermark based detector. The extensive simulations prove that this attack could be used to deceive the system even with the presence of a dynamic watermark.

Smart vehicles can be more adaptive to the road condition by exchange information between the vehicle. They can avoid traffic congestion, dangerous obstacles, even traffic accident earlier. This technology is closely related to the safety of the driver, therefore it must receive special attention. V2V communication has the potential to threaten interference and even attacks. There have been many studies that have focused on finding solutions to deal with these disorders. The first step is to strengthen the system's ability to detect attacks on V2V. On the other hand, the development of Machine Learning (ML) looks very promising to support this goal. In the proposed scheme, a 2-Step Prediction for detecting attackers is used. This system is using two classifiers ML from two modified training datasets. We show that the proposed scheme can improve the attack detection performance compared to one detection step.

Due to the sensitivity of the information exchanged in Vehicle to Vehicle (V2V) and Vehicle to Infrastructure (V2I) communication, generating secret keys is critical to secure these communications. As nature is open access, distributed keys are more vulnerable to attacks in the vehicular environment. Physical layer key generation methods using wireless channel characteristics show promise in preventing such attacks, generating keys independently, and removing the need for distribution. In this paper, we present a novel key generation approach in a real vehicular environment based on Channel State Information (CSI), including a new algorithm for key bit extraction. We implemented our algorithm using USRP B210 Software-Defined Radios (SDR) and the industry-standard V2X communication protocol: IEEE 802.11p. The proposed key generation protocol uses the CSI values of each sub-carrier as a source of randomness, from which bits are extracted using a new QAM demodulator quantizer (QAM-Dem-Quan). We compared our technique to state-of-the-art Received Signal Strength (RSS)-based approaches, and show that our method achieves better performance. Moreover, we reached a min-entropy of approximately 70% for the generated keys and a key generation rate of less than$150~\mu \text{s}$/key for key lengths ranging from 16 to 128 bits.

Since the vehicular environment is highly mobile, the transmitted messages are affected by the wireless channel effect. This makes channel estimation one of the important tasks in Vehicle-To-Everything (V2X) communications. In this paper, we propose a novel Multiple Sequential Constraint Removal (MSCR) algorithm of channel estimation, dedicated to V2X communications. This method is performed on a vehicular doubly selective channel, where the communication system is based on the Orthogonal Frequency Division Multiplexing (OFDM) waveform, with a high order modulation (64-QAM). The proposed approach is compared to state-of-the-art Least square (LS), minimum mean square error (MMSE) estimators and the iterative Sequential Constraint Removal (SCR) algorithms. It shows that it achieves better performance, where we can get a small symbol error rate (SER) on highly mobile scenarios (Highway with non-line of the sight), with a latency time that meets with the V2X communication requirements.

Inter-vehicular communication is a major research field in the intelligent transportation systems (ITS) industry. It has been increasingly growing due to recent advances in mobile and wireless communication technologies. This paper aims to present a novel cross layer channel estimation technique for inter-vehicular communication based on Bayesian network theory. It proposes a multi-criteria estimation method of the Orthogonal Frequency Division Multiplexing OFDM in the 802.11p communication. The proposed method seeks to enhance the way the standardized and initial estimation method proposed in the V2V standard interact with its environment. The paper introduces two estimation-based pilot subcarrier techniques. The first technique considers re-arranging the initial position of pilot subcarriers, and the second technique adds two supplementary subcarriers. A Bayesian channel estimation technique is proposed wherein a decision-aided algorithm starts by estimating the impact of the information to be transmitted and then proceeds by assessing the error rate of the previously transmitted data while taking the quality of transmission into account. The results show that the proposed system responds well to instances involving degradation in the communication environment..

The equipment of vehicles with wireless communication capabilities is expected to be the key to the evolution to next generation intelligent transportation systems (ITS). The IEEE community has been continuously working on the development of an efficient vehicular communication protocol for the enhancement of Wireless Access in Vehicular Environment (WAVE). Vehicular communication systems, called V2X, support vehicle to vehicle (V2V) and vehicle to infrastructure (V2I) communications. The efficiency of such communication systems depends on several factors, among which the surrounding environment and mobility are prominent. Accordingly, this study focuses on the evaluation of the real performance of vehicular communication with special focus on the effects of the real environment and mobility on V2X communication. It starts by identifying the real maximum range that such communication can support and then evaluates V2I and V2V performances. The Arada LocoMate OBU transmission system was used to test and evaluate the impact of the transmission range in V2X communication. The evaluation of V2I and V2V communication takes the real effects of low and high mobility on transmission into account. The yielded results will help us to validate previous Matlab simulations of the IEEE 802.11p transmission system.

With the growth of embedded components (GPS, radar, cameras...), road vehicle becomes more and more intelligent and reassuring. The concept of the intelligent vehicle is to make it able to analyze, individually and independently, the external data before transmitting them to the driver. In recent years, ideas for studies and research programs on this concept have proliferated in different environments. The cars now have tools to control speed, manage traffic and ensure the avoidance collision task. A radar, that allows to detect obstacles in the road environment and to alert the driver, is one of the very powerful devices, even essential to ensure the safety of road users. This device has the advantage of being effective anytime (fog, rain...). The arrival of Ultra Wide Band (UWB) technology for radar application allows the development of compact and relatively cheap sensors. Dedicated to military applications, radar devices using UWB are now a good tool of detecting obstacles for many applications in life everyday. These sensors could be used to measure distances and positions with greater resolution than existing radar devices or to obtain images of objects buried underground or placed behind surfaces. This paper is focused on the study of UWB radar signatures to identify the type of obstacles.

This article discusses the design of an application specific MPSoC architecture dedicated to multiple target tracking (MTT). This application has its utility in driver assistant systems, more precisely in collision avoidance and warning systems. An automotive-radar is used as the front end sensor in our application. The article examines the tradeoffs that must be taken into consideration in the realization of the entire MTT application in an embedded system. In our implementation of MTT, several independent parallel tasks have been identified and mapped onto a multiprocessor architecture to ensure the deadlines imposed by the application. Our study demonstrates that the joint utilization of reconfigurable circuits (namely FPGA) and MPSoC, facilitates the development of a flexible and efficient MTT system.

In order to improve exploitation in guided transport networks, many approaches have been investigated for the development of tools or systems able to provide needed information. Ours laboratories designed an original radar system that allows inter-vehicle communication. This system called CODIREP - communication detection and identification of broken-down trains - is based on the principle of a co-operative radar using a transponder inside targets. It is made of a transmitter/receiver couple, which equips respectively the front and the rear of two successive vehicles. It uses a numerical correlation receiver and has a broad band of about 100MHz which could be exploited to establish high data flow communications. A preliminary prototype was realized using a technical solution to multiplex communication data and localization code. The tests in free space and tunnel show that a range of 800 m in free space and 700 m in tunnel with a location precision of 1.5 m and a data rate of about 1.6 Mbps could be reached. This remains insufficient for some applications that require more resources. With the system, we intend to reach more than 5 Mbps. The aim of this work is to improve this system by proposing another original coding technique developed in order to increase the data flow rate. Simulations will be executed to evaluate the system's performances in terms of data flow, bit error rate (BER), computing time and complexity.

This paper describes the real time processing unit used for an anticollision road radar system. This radar based on a numerical correlation between the transmitted signal and the received signal is under development. The signal uses orthogonal codes to ensure a multiple access communication between all vehicles in near area. In this paper, a real time processing unit associated to an original anticollision radar is presented. The studied radar is based on spreading spectrum coded radar waveforms at 76-77 GHz and a numerical correlation receiver. This sensor associated to other sensors like Lidar and Camera will be used on-board vehicles for more safety on road. The studied receiver computes the numerical crosscorrelation between the received signal and a replica of the transmitted code to allow an optimal detection. The appropriate coding and processing have been used to implement a laboratory radar mock-up. The real time processing is tested in order to show their performances and disadvantages when applied to obstacles detection. The main idea is to achieve an efficient real time detection using a simple and low cost system.