repository·master·Indexed 25 days ago
https://github.com/sherlockchou86/videopipeA C++ framework for video analysis and structuring featuring a plugin-oriented pipeline architecture. It supports the integration of computer vision models for object detection, tracking (Deep SORT, SORT), and behavior analysis. The framework includes a variety of nodes for inference (OpenCV DNN, TensorRT, PaddleInference), on-screen display (OSD), data brokering (Kafka, UDP, JSON), and recording, with support for multi-channel processing and hardware acceleration via CUDA and TensorRT.
nlohmann/json library is a header-only JSON library used in VideoPipe to parse and dump JSON payloads when interacting with REST APIs from mLLM (multimodal Large Language Model) services. It provides a syntax that is easy to use, similar to working with JSON in Python.The trt_vehicle component provides several computer vision capabilities for vehicle analysis:
bhtsne library is an implementation of the t-SNE (t-distributed Stochastic Neighbor Embedding) algorithm. It is used to reduce high-dimensional features into 2D or 3D representations, typically for visualization purposes on a screen.kissnet is a lightweight, header-only socket library used specifically for logging and message brokering via the UDP protocol.https://github.com/ReneNyffenegger/cpp-base64. In the context of VidepPipe, it is specifically used for encoding images into base64 strings to facilitate interaction with multi-modal Large Language Models (mLLM). The library has been modified to merge .cpp and .h files together for easier integration.Node types are connected like a pipeline to build video analysis applications. It is designed for low dependency, high portability, and ease of use in scenarios like video structuring, image search, face recognition, and traffic/security behavior analysis.nodes/track/deep_sort module implements the Deep SORT (Simple Online and Realtime Tracking with a Deep Association Metric) algorithm for object tracking within the videopipe pipeline.vp_record_node is a middle node in the videopipe pipeline used to record video and images. It saves the captured media to the local disk once the recording process is finished.
Because it operates asynchronously, the recording process does not block the rest of the pipeline execution.
nodes/track/sort module provides a sort algorithm specifically designed for tracking tasks within the videopipe pipeline.