In a world where cities are evolving into increasingly connected and intelligent ecosystems, managing urban infrastructure has become a critical challenge.
Optimizing traffic flow, enhancing public safety, and designing urban spaces that meet citizens' needs require a deep understanding of the urban environment. While these challenges are complex, they also present a unique opportunity: leveraging advanced technologies, such as artificial intelligence and computer vision, to reimagine tomorrow’s cities and significantly improve their efficiency.
Object tracking serves as a cornerstone of computer vision systems. This technique aims to associate detected objects across successive frames, enabling accurate counting and in-depth analysis of their movements, such as speed, trajectory, and deformation. Today, this field is primarily dominated by two major approaches:
1. Detection-Based Tracking
This method relies on detection algorithms at each frame, followed by the association of detected objects between frames. Popular algorithms like SORT (Simple Online and Realtime Tracking), coupled with detection models such as YOLO, are frequently used. This solution stands out for its speed and accuracy in optimal conditions. However, it shows its limitations when faced with challenges such as occlusions, erratic movements, densely populated environments, or even object deformation, abrupt directional changes, and dilation effects.
2. Feature-Based Tracking
This approach relies on the use of specific features, such as key points or regions of interest, to track moving objects. Techniques like Optical Flow offer precise tracking but also have notable limitations, such as:
Sensitivity to lighting variations: Changes in lighting can disrupt the tracking.
Object deformation: Alterations in shape or appearance complicate the process.
These two tracking methods, while effective, quickly show their limits when combined with traditional counting algorithms based on a single-target per object, as well as in low frame rate situations (~2 frames per second). In tracking algorithms, it is important to note that the min-hits parameter, which determines the threshold for tracking an object, is closely tied to the frame rate (FPS). In the SORT method, this value is close to 3 to ensure optimal tracking and reduce noise (eliminating false positives). In the case of a single-target algorithm, if min-hits exceed the frame rate value, the counting error rate can exceed 85%, especially in complex and varied environments such as densely populated areas or regions with heavy traffic.
To overcome these limitations, the multi-tracking points approach proposes tracking multiple targets per object, ensuring at least one target is counted. This helps to minimize the error percentage as much as possible.
Towards More Efficient Counting with Multi-Tracking Points
VizioSense has developed a multi-point tracking algorithm, offering superior efficiency compared to traditional methods that typically track only a single point or region per object. How does it work?
1. Tracking Multiple Points per Object
By assigning multiple points to an object, VizioSense's algorithms provide an optimized reconstruction of trajectories, even in complex situations involving partial occlusions, object intersections, or unpredictable movements.
Unlike traditional methods, which are limited to a single point or region, this multi-point approach allows precise tracking of objects, even when their appearance changes or they are temporarily hidden. Each object is represented by a series of strategically positioned points, ensuring reliable and continuous tracking despite variations in position, lighting, or perspective. This innovative approach significantly improves counting accuracy and reduces errors caused by factors such as:
Occlusions: Temporarily hidden objects remain identifiable through the other tracked points.
Dense environments: The ability to manage multiple close objects prevents trajectory association errors.
Dynamic variations: Changes in lighting, angle, or perspective do not affect the overall tracking of objects.
The green points represent potential targets. Each point is a strategic reference to ensure precise and robust tracking. The red rectangles frame the detected objects, while the trajectories illustrate the algorithm's ability to associate the points to maintain tracking consistency over time and space. The algorithm offers exceptional resilience in dense environments or dynamic urban scenarios, like detecting parking spots.
2. Error Reduction
The multi-point approach automatically corrects detection errors by intelligently associating trajectories. It also reduces noise generated by detection, limiting false positives that could mistakenly be taken as targets to track. For example, if a vehicle is temporarily obscured by another, the algorithm continues tracking it by relying on the other associated points, ensuring continuity and increased accuracy in tracking.
It is also essential to note that a multi-point tracking system with a frame rate (FPS) of only 2 frames per second provides performance equivalent to a single-point tracking system that requires a frame rate of 5 to 6 frames per second.
3. Robustness in Varied Environments
The multi-point tracking adapts to changing conditions, such as lighting variations, complex viewing angles (sensor positioning), or challenging weather conditions (rain, fog, snow, sun). This makes it ideal for outdoor use cases, on roads, or in public spaces.
Conclusion
Through its innovations in multi-tracking, VizioSense has revolutionized the counting of vehicles and people, addressing the challenges posed by modern urban environments. Its technology not only improves accuracy but also enhances robustness and reliability in difficult conditions, achieving an accuracy rate of > 98.5%, which corresponds to an error rate of < 1.5%.
As smart cities continue to grow, VizioSense's algorithms pave the way for even more ambitious applications. By combining expertise in artificial intelligence and computer vision, VizioSense is positioning itself as a key player in this rapidly evolving field. From enhancing passenger flow management in airports and optimizing crowd control in train stations to improving traffic monitoring on busy roadways, VizioSense's technology offers practical solutions for a wide range of urban challenges.