- [email protected]
- Mon - Sat: 8.00 am - 7.00 pm
Your innovative IT execution crew! Hire Us
Multicam human tracking AI
Partner: Vietnam National University of Engineering AI Research Lab (UET AI Research Lab)
Project Overview: Our researchers has collaborated with the Vietnam National University of Engineering AI Research Lab (UET AI Research Lab) to develop a state-of-the-art AI Multicam Tracking System. This innovative project focuses on mapping humans captured on camera to a map in real-time. By leveraging advanced computer vision, AI technologies, MLOps techniques, and powerful computing resources like Nvidia DGX, the multicam tracking system delivers accurate and efficient tracking capabilities, suitable for a wide range of applications in security, retail, and urban planning.
Project Objectives:
- Develop a multicam tracking system that can accurately map humans captured on camera to a real-time map.
- Collaborate with UET AI Research Lab to leverage their expertise in AI and computer vision technologies.
- Implement MLOps techniques to streamline the development, training, and deployment of AI models.
- Utilize Nvidia DGX for accelerated training and processing of the tracking system.
- Create a versatile system that can be used in various industries, such as security, retail, and urban planning.
Techniques and Technologies Used:
- Three main module:
- Multiple Object Tracking (MOT) module: Utilize object detection, re-identification models, pose estimation and motion estimation algorithms to track staff members in each camera. This module combines feature representation, spatial information, and motion direction to achieve accurate tracking results.
- Message Broker: We use kafka for streaming data between MOT and MTMC module
- Multiple Target Multiple Camera (MTMC) module: Aggregate the tracking results obtained from multiple cameras. This module employs spatial and temporal rules to track staff members across all cameras. Additionally, it incorporates enhanced features derived from the analysis of multiple camera views to improve tracking accuracy and robustness
- To achieve near real-time performance:
- The MOT module is implemented using Deepstream, a framework specifically designed for efficient video stream processing.
- All deep learning models are converted to TensorRT
- Project Outcomes and Benefits:
- Performance Metrics:
- Demonstrated a commendable ~90 HOTA score in single-camera tracking scenarios, indicating accurate and reliable tracking results.
- Achieved a competitive ~70 HOTA score in multiple-camera tracking scenarios, showcasing the system’s ability to effectively handle complex tracking challenges across different camera views.
- System Efficiency:
- Exhibited efficient performance by maintaining a solid 20 frames per second (FPS) processing rate while simultaneously handling the data from 10 cameras.
The AI Multicam Tracking System offers several advantages, including:
- Enhanced security: The system can be used for monitoring and securing premises, tracking suspicious activity, and providing real-time alerts.
- Improved retail analytics: The system enables retailers to gain insights into customer behavior, foot traffic patterns, and store layout effectiveness.
- Urban planning and management: By tracking and analyzing human movement, the system can provide valuable data for optimizing public spaces, transportation networks, and infrastructure.
- Scalability: The system can accommodate various camera configurations and can be scaled up or down to suit different applications and environments.
- Real-time mapping: The ability to map humans captured on camera to a real-time map allows for prompt decision-making and efficient resource allocation.
- Streamlined AI development: The implementation of MLOps techniques ensures efficient collaboration between teams, accelerating the development, training, and deployment of AI models.
- Accelerated performance: Utilizing Nvidia DGX for training and processing enhances the system’s performance and enables faster, more accurate tracking.
The AI Multicam Tracking System is a testament to the successful collaboration between our researchers and the UET AI Research Lab. This innovative solution, backed by cutting-edge technologies and MLOps practices, has the potential to transform industries and improve the way organizations monitor and analyze human movement.
