Traffic analytics configuration guide
This document outlines recommended camera installation and configuration guidelines for achieving accurate people and vehicle counting results. Adhering to these best practices will optimize the performance of your video analytics system and minimize counting errors.
Camera resolution:
Recommended camera resolution is 2 MP, 25 FPS.
VMS traffic analytics configuration recommendations:
Confidence level: 60-80
Frame skip: 2-3
IMPORTANT! Using a GPU is highly recommended when running the traffic analytics application. Running on a CPU may lead to reduced accuracy and increased resource consumption.
Key Considerations:
· For accurate detection, it's essential that a person's entire body fully enters the entrance zone and remains there for 1-2 seconds before proceeding to the exit zone, also remaining there for 1-2 seconds. Clear visibility of the body within both zones is also critical.
· Zone Configuration within the VMS: Proper configuration of "in" and "out" zones within the VMS is critical for accurate counting. These zones should be wide enough to allow the system to reliably track objects as they enter and exit.
· Object Dwell Time within Zones: A crucial factor is the "dwell time" – the duration an object remains entirely within a designated zone. Ideally, an object (person or vehicle) should be fully contained within each zone for at least 1-2 seconds. This allows the system to capture sufficient data points for accurate tracking and classification.
· Mitigating Sudden Appearance and Rapid Transitions: The system's ability to accurately count objects is compromised when individuals or vehicles appear suddenly within a zone or transition rapidly between zones. Such scenarios can prevent the system from establishing a reliable track, leading to missed counts. Minimize these occurrences by carefully selecting camera placement and adjusting zone parameters.
· Environmental Factors: Consider environmental factors that can influence accuracy. For example, shadows, rain, or snow can affect the performance of the analytic.
Illustrative Examples:
The following examples demonstrate optimal and suboptimal camera views for people counting:
