How Hugo Improved Object Detection for Autonomous Surveillance Drones
The Client
The Client is a security company that leverages human expertise and technology to enhance indoor surveillance. Founded in 2018, it provides 24/7 monitoring for offices, industrial sites, and warehouses, helping organizations strengthen security and facility management. Its system detects critical events like water leaks, fires, power outages, and temperature anomalies, ensuring proactive response and operational efficiency.
The Challenge
The Client's autonomous security drones rely on AI and fisheye lenses to deliver 360° surveillance. Ensuring accurate object detection in complex environments proved challenging due to:
Object Detection Complexity: Doors lacked distinctive features, varied in size, and were difficult for the AI to recognize.
Lens Distortion: The fisheye effect altered object proportions, complicating identification.
Environmental Variability: Changing lighting conditions affected detection accuracy.
Motion & Static Detection: The system needed to differentiate between stationary objects (doors, windows) and moving subjects (people, vehicles).
"A good model that can perform reliable detections is critical to our entire operation," said Chen A., R&D Lead.
The Solution
Hugo implemented a targeted strategy to improve the Client's AI accuracy in security surveillance.
Specialized Security Annotation Expertise
Hugo assembled a dedicated team with specific training in security imagery and surveillance systems. These specialists developed an understanding of the unique visual challenges presented by the Client's drone-captured footage, including the fisheye lens distortion and varying lighting conditions typical in security applications.
Enhanced Annotation Quality Process
Hugo established a rigorous quality framework specifically designed for security applications:
Implemented multi-stage review protocols where annotations underwent initial labeling, peer verification, and final expert validation
Created specialized guidelines for consistent identification of challenging objects like doors and windows
Developed annotation techniques to account for lens distortion effects on object appearance
Applied precision bounding box methodologies to ensure complete object capture
Comprehensive AI Training Improvement
Hugo’s annotation process directly improved the Client’s machine-learning model by:
Improving training data quality for enhanced object recognition.
Reducing false positives and negatives through consistent labeling.
Implementing Human-in-the-Loop processes to ensure complete element labeling.
Establishing intensive quality assurance measures to maintain data integrity.
The Results
40% reduction in false positives, minimizing unnecessary alerts.
35% improvement in object recognition, enhancing detection of doors and key structures.
98.5% annotation acceptance rate, ensuring high data accuracy.
100% on-time delivery, meeting strict project timelines.
30% faster model training, accelerating deployment.
Operational Impact
Hugo’s high-quality annotations improved the Client’s security drone AI, resulting in accurate anomaly detection, fewer false alarms, and a successful product launch. The partnership also ensured GDPR compliance by excluding sensitive data while maintaining diverse, high-quality training inputs.
"Hugo felt like an extension of our team, not just an outsourcing vendor. Their expertise in visual data annotation provided invaluable insights to improve our ML models." – Chen A., R&D Lead.
