SOVABench: A Vehicle Surveillance Action Retrieval Benchmark for Multimodal Large Language Models

Automatic identification of events and recurrent behavior analysis are critical for video surveillance. However, most existing content-based video retrieval benchmarks focus on scene-level similarity and do not evaluate the action discrimination required in surveillance. To address this gap, we introduce SOVABench (Surveillance Opposite Vehicle Actions Benchmark), a real-world retrieval benchmark built from surveillance footage and centered on vehicle-related actions. SOVABench defines two evaluation protocols (inter-pair and intra-pair) to assess cross-action discrimination and temporal direction understanding. Although action distinctions are generally intuitive for human observers, our experiments show that they remain challenging for state-of-the-art vision and multimodal models. Leveraging the visual reasoning and instruction-following capabilities of Multimodal Large Language Models (MLLMs), we present a training-free framework for producing interpretable embeddings from MLLM-generated descriptions for both images and videos. The framework achieves strong performance on SOVABench as well as on several spatial and counting benchmarks where contrastive Vision-Language Models often fail. The code, annotations, and instructions to construct the benchmark are publicly available.

Visual Context-Aware Person Fall Detection

As the global population ages, the number of fall-related incidents is on the rise. Effective fall detection systems, specifically in healthcare sector, are crucial to mitigate the risks associated with such events. This study evaluates the role of visual context, including background objects, on the accuracy of fall detection classifiers. We present a segmentation pipeline to semi-automatically separate individuals and objects in images. Well-established models like ResNet-18, EfficientNetV2-S, and Swin-Small are trained and evaluated. During training, pixel-based transformations are applied to segmented objects, and the models are then evaluated on raw images without segmentation. Our findings highlight the significant influence of visual context on fall detection. The application of Gaussian blur to the image background notably improves the performance and generalization capabilities of all models. Background objects such as beds, chairs, or wheelchairs can challenge fall detection systems, leading to false positive alarms. However, we demonstrate that object-specific contextual transformations during training effectively mitigate this challenge. Further analysis using saliency maps supports our observation that visual context is crucial in classification tasks. We create both dataset processing API and segmentation pipeline, available at https://github.com/A-NGJ/image-segmentation-cli.

Unified Physical Threat Monitoring System Aided by Virtual Building Simulation

With increasing physical threats in recent years targeted at critical infrastructures, it is crucial to establish a reliable threat monitoring system integrating video surveillance and digital sensors based on cutting-edge technologies. A physical threat monitoring solution unifying the floorplan, cameras, and sensors for smart buildings has been set up in our study. Computer vision and deep learning models are used for video streams analysis. When a threat is detected by a rule engine based on the real-time analysis results combining with feedback from related digital sensors, an alert is sent to the Video Management System so that human operators can take further action. A physical threat monitoring system typically needs to address complex and even destructive incidents, such as fire, which is unrealistic to simulate in real life. Restrictions imposed during the Covid-19 pandemic and privacy concerns have added to the challenges. Our study utilises the Unreal Engine to simulate some typical suspicious and intrusion scenes with photorealistic qualities in the context of a virtual building. Add-on programs are implemented to transfer the video stream from virtual PTZ cameras to the Milestone Video Management System and enable users to control those cameras from the graphic client application. Virtual sensors such as fire alarms, temperature sensors and door access controls are implemented similarly, fulfilling the same programmatic VMS interface as real-life sensors. Thanks to this simulation system's extensibility and repeatability, we have consolidated this unified physical threat monitoring system and verified its effectiveness and user-friendliness. Both the simulated Unreal scenes and the software add-ons developed during this study are highly modulated and thereby are ready for reuse in future projects in this area.