Release Notes: Image-Based Video Search#
Details about the changes, improvements, and known issues in this release of the application.
Version 1.3.0#
New
Qualified on the Intel® Core™ Series 3 processor (Wildcat Lake).
Deprecated EMF deployment packages.
Improved
Bug fixes to improve UI/UX.
Version 1.2.0#
New
Qualified on the Intel® Core™ Ultra Series 3 processor.
Changed the person-vehicle-bike-detection-2004 model to the yolov11s model.
Improved
Consumed the latest DL Streamer Pipeline Server 2026.0.0 image. Ubuntu24 variant of the image is the default now.
Image-based Video Search now supports the NPU device.
Version 1.1.0#
Added Nginx reverse proxy server, updated npm packages and python packages
Fixed high coverity issues and added healthchecks for compose and helm deployment
DL Streamer and DL Streamer Pipeline Server versions upgraded to 2025.2
Version 1.0.1#
Release Date: 2025-08-20
DL Streamer version upgraded to 2025.1.2
DL Streamer Pipeline Server version upgraded to 3.1.0
Bug fixes
Version 1.0.0#
Release Date: 2025-03-31
MediaMTX: This third-party microservice simulates remote video cameras placed in various geographical locations. The service can replay videos from recorded files and publish them as live video streams in different formats. The RTSP stream is sent to DL Streamer Pipeline Server microservice for extraction of feature vectors. The WebRTC stream is useful for displaying live video streams in this web application.
Deep Learning Streamer Pipeline Server: This microservice operates either in streaming mode (video analysis) or image ingestion mode (video search). During the video analysis phase, it consumes RTSP streams from simulated remote cameras and performs object detection and feature extraction tasks. The resulting predictions are published as metadata to an MQTT Broker. During the video search phase, the microservice returns feature vectors for submitted images.
Feature Matching: This microservice interacts with DL Streamer Pipeline Server and Vector DB. During the video analysis phase, it inserts feature vectors (image embeddings) generated by the Pipeline Server into Vector DB. Additionally, it stores frames that are useful for displaying in search results when a user searches for an object. During the video search phase, the microservice queries the Pipeline Server to get feature vectors for the user-provided image and then queries Vector DB to find the matching frame(s) from video streams processed in the analysis phase.
MQTT: This third-party microservice transfers a stream of metadata between DL Streamer Pipeline Server and Feature Matching during the video analysis phase.
MilvusDB: This third-party microservice stores feature vectors in a Vector DB and provides a vector search to find the closest match between a user-provided image and previously processed video frames.