This paper presents a task-adaptive open-set detection framework that preserves zero-shot performance while incorporating task-specific adaptations for enhanced visual understanding. Our method integrates a frozen zero-shot detector with a learnable, task-specific adaptor module, and employs a token-level conditional inference mechanism using prompt-based feature masking. This approach selectively combines features from the pre-trained zero-shot model and the adapted module within a single forward pass, allowing both general and task-specific representations to contribute effectively. Unlike conventional full fine-tuning that transforms an open-set detector into a closed-set detector, our design maintains the inherent open-set capabilities, thereby mitigating overfitting to task-specific biases. Experimental results on the IHP and VFP290K datasets demonstrate that our method outperforms existing techniques in fallen person detection, underscoring its robustness and practical applicability.

Precise localization encompassing diverse indoor spaces is the key to immersive interaction services. In practice, indoor localization often undergoes blind spots as RF is easily blocked by everyday objects ranging from concrete walls, metallic shelves, and partitions to electronics and appliances. This paper presents SuperSight, an NLOS localization for mmWave backscatter that, for the first time, achieves NLOS (non-penetrable) localization without multipath environment profiling/manipulation. The key insight of SuperSight is uniquely exploiting the mmWave features of highly directional and specularly reflected multipath in combination with the triangular tag array to yield sub-cm NLOS localization accuracy over 8 m range - an order of magnitude performance enhancement compared to the competitors. Circularly polarized, 77GHz retro-reflective tag ensures high precision and robustness across diverse reflectors and tag orientations. The prototype was evaluated across six different reflector materials (including metal, concrete, and plaster) and demonstrated in the corridor and office space to reveal x, y, z position accuracy of up to (metal reflector) 5.7 mm, 5.5 mm, 7.7 mm at 8 m range, with Yaw, Pitch, Roll accuracy of 0.22, 0.28, 0.1 degrees respectively.

Accurate and scalable localization is one of the keys to pervasive interaction with the Internet of Things. mmWave backscatter possesses great potential toward this goal - The abundant bandwidth of mmWave enables high-precision localization, and low-cost and ultra low-power backscatter tags enable massive deployment with minimum deployment cost and maintenance efforts. We present Hawkeye, a new mmWave backscatter that offers (i) submillimeter localization accuracy (ii) at over 2 m range, (iii) while consuming 2.25 μW power. At the heart of our design is the new Hawkeye super-resolution, which exploits the interplay between the tag FSK and FMCW radar to improve the localization performance by ×60 over conventional FMCW radar (i.e., c/2BW). Hawkeye readers were implemented on commodity 77 GHz radars and the tags were prototyped on PCB.

Accurate localization of a large number of objects over a wide area is one of the keys to the pervasive interaction with the Internet of Things. This paper presents Hawkeye, a new mmWave backscatter that, for the first time, offers over (i) hundred-scale simultaneous 3D localization at (ii) subcentimeter accuracy for over an (iii) hectometer distance. Hawkeye generally applies to indoors and outdoors as well as under mobility. Hawkeye tag's Van Atta array design with retro-reflectivity in both elevation and azimuth planes offers 3D localization and effectively suppresses the multipath. Hawkeye localization algorithm is a lightweight signal processing compatible with the commodity FMCW radar. It uniquely leverages the interplay between the tag signal and clutter, and leverages the spectral leakage for fine-grained positioning. Prototype evaluations in corridor, lecture room, and soccer field reveal 7 mm median accuracy at 160 m range, and simultaneously localize 100 tags in only 33.2 ms. Hawkeye is reliable under temperature change with significant oscillator frequency offset. Demo video: https://tinyurl.com/4zkwxatu

Massive connectivity is a key to the success of the Internet of Things. While mmWave backscatter has great potential, substantial signal attenuation and overwhelming ambient reflections impose significant challenges. We present OmniScatter, a practical mmWave backscatter with an extreme sensitivity of -115 dBm. The performance is theoretically comparable to the popular commodity RFID EPC Gen2 (900 MHz), and is empirically validated via evaluations under various practical settings with abundant ambient reflections and blockages - e.g., In an office where a tag is locked in a wooden closet 6m away, as well in libraries and retail stores where a tag is placed across two rows of metal shelves. At the heart of OmniScatter is the new High Definition FMCW (HD-FMCW), which interplays with the tag (FSK) signal to disentangle the ambient reflections from the tag signal in the frequency domain, essentially offering immunity to ambient reflections. To further support practical deployment, OmniScatter offers coordination-free Frequency Division Multiple Access (FDMA) that effortlessly scales to thousands of concurrent tags. The readers were built on commodity radars and the tags were prototyped on PCB. The trace-driven evaluation demonstrates concurrent communication of 1100 tags with the BER < 1.5%, paving a pathway towards practical mmWave backscatter for everyday and anywhere use.

In many visual applications, we often encounter people with irregular postures, such as lying down. Many approaches adopted two-step methods to handle a person with irregular postures: 1) person detection and 2) posture prediction based on the detected person. However, it is challenging to detect irregular postures because the existing detectors were trained with datasets consisting of most upright postures. Therefore, we propose a new Irregular Human Posture (IHP) dataset to handle various postures captured from real-world surveillance cameras. The IHP dataset provides sufficient annotations to understand the posture of person, including segmentation, keypoints, and postural states. This paper also provides two baseline net-works for postural state estimation of the people trained on the IHP dataset. Moreover, we show that our baseline networks effectively detect the people with irregular postures that may be in an urgent situation in a surveillance environment.

Moving object detection in a video plays an important role in many vision applications. Recently, moving object detection using appearance modeling based on a convolutional neural network has been actively developed. However, the CNN-based methods usually require the user's supervision of the first frame so that it becomes highly dependent on the training dataset. In contrast, the method of finding a foreground, which models a background occupying a large proportion in an image, can detect a moving object efficiently in an unsupervised manner. However, existing methods based on background modeling in a pan-tilt-zoom (PTZ) camera suffer many false positives or loss of moving objects due to the estimation error of camera motion. To overcome the aforementioned limitations, we propose a moving object detection method for a PTZ camera through two background models. In an unsupervised way, our method builds the two background models that have different roles: 1) a coarse background model for detecting large changes, and 2) a fine background model for detecting small changes. In more detail, the coarse background model builds a block-based Gaussian model, and the fine model builds a sample consensus model. Both models are adaptively updated according to the estimated camera motion in the video recorded by a PTZ camera. Then, each foreground result from two background models is incorporated to fill the moving object region. Through experiments, the proposed method achieves better performance than the state-of-the-art methods and operates in real-time without parallel processing. In addition, we showed the effectiveness of the proposed model through improved results of moving object detection through combination with the latest supervised method.