NexVDO SDK

Analysis

Video content, biometric, and behavior-analysis APIs

NexVDO SDK combines video input, recording, streaming, and analysis for cloud or edge deployments. Teams can choose the architecture based on latency, privacy, connectivity, and operating requirements. The analysis module groups its capabilities into Video Content Analysis, Biometric Recognition, and Behavior Analysis; model performance must be validated with the target cameras, data, and environment.

Video Content Analysis provides image segmentation, object detection, classification and counting, feature matching, and text recognition. Example workflows include people or vehicle counting, cross-camera search, industrial inspection, license-plate recognition, and medical-image processing. The appropriate model and threshold depend on the application and validation data.

Biometric Recognition supports 1:N face matching, profiles turned to approximately 120 degrees, 68 three-dimensional facial landmarks, and six expression labels: calm, surprise, happiness, anger, disgust, and fear. YUAN reports a 99.8 recognition-rate result and participation in NIST FRVT; teams should confirm the test version, dataset, threshold, and current applicability before using that figure. Deployments also need consent, privacy, bias, and legal review.

Behavior Analysis tracks 17 body keypoints across a sequence of frames. Example workflows include 1080p240 motion review for sports, fall-event review in care settings, and classroom activity analysis. These systems can flag events for review; they do not establish safety, health, or learning outcomes without application-specific evidence.

The SDK supports NVIDIA and Intel GPU paths and works with YUAN capture, recording, and streaming modules. Developers should measure inference speed, image quality, memory use, and model accuracy on the intended hardware before deployment.