Multi-Camera Vehicle Tracking Using Edge Computing and Low-Power Communication

Area of Science:

• Computer Vision
• Artificial Intelligence
• Embedded Systems

Background:

• Traditional large-area visual vehicle tracking relies on multiple cameras and complex algorithms, often requiring data transmission to powerful workstations.
• This centralized processing leads to high data volumes, significant communication bandwidth demands, and potential privacy concerns.

Purpose of the Study:

• To develop and evaluate deep learning-based detection and tracking algorithms that can operate directly on embedded camera systems.
• To reduce data transmission requirements and communication bandwidth for multi-camera vehicle tracking.

Main Methods:

• Implementation of dedicated deep learning algorithms for vehicle detection and tracking directly on camera embedded systems.
• Development of a low-power, short-range communication protocol for inter-camera data exchange.
• Integration of algorithms and communication for decentralized multi-target, multi-camera tracking.

Main Results:

• Significant reduction in data stream from cameras.
• Reduced communication bandwidth requirements.
• Successful implementation of multi-camera tracking directly within camera systems, evaluated across diverse environmental conditions.

Conclusions:

• On-camera deep learning processing offers an efficient solution for large-area visual vehicle tracking.
• The proposed system minimizes data transmission, enhances privacy, and enables flexible communication options.
• The solution is suitable for applications like parking management and intersection monitoring.