Video surveillance systems are gaining more popularity these days and are commonly used for security, traffic analysis, agriculture control systems etc.
With our expertise in computer vision and web development, we created the real-time vehicles tracking app that can be easily modified to tracking other objects (people, animals, plants etc.) and serving various business scenarios. It could be used in different modes: from the on-demand single frame detector to the continuous stream analyzer and tracker.
- Counting vehicles that pass by (independently in any user-defined directions)
- Detecting different kinds of vehicles (cars, trucks, motorcycles)
- Different service working modes:
- single image processing over REST API with blazing low latency of 450ms
- smooth and adaptive live-streaming in peer-to-peer technology with end-to-end latency ≈ 2 seconds
- reliable live-streaming in server-client architecture designed to support various user’s devices
- Possibility to process various stream sources, like a dedicated camera (e.g. IP-Camera) or a user’s web camera
A deep learning-based detector along with a multi-object tracking system is used for video processing on our server. Real-time two-way communication allows continuous analysis of incoming frames and sending annotations to web clients, which lead to smooth live-streaming experience.
The system accepts various stream sources (IP or web camera, video files etc.) and allows multiple clients to be connected for the annotated video stream. The solution is built using microservices as isolated Docker containers.