A CCTV provider needed help detecting shoplifting. Learn how Aya created a real-time system for detecting theft and warning behaviors.
Overview
Shoplifting is a major issue, with over 200 million incidents reported annually in the U.S. alone—equating to more than 500,000 cases a day. Globally, shoplifting results in over $120 billion in losses each year. To combat this problem, many companies are turning to technology for innovative solutions. A leading CCTV analytics provider sought to develop a model that could detect shoplifting in real-time, and they needed precise labelling of shoplifting events for training this model.
The Challenge
Aya Data was tasked with labelling 5,000 hours of CCTV footage to identify shoplifting events. This involved:
– Labelling Individuals: Each person in every frame of the footage had to be labelled with precise bounding boxes.
– Marking Theft Events: Specific key points were added to indicate the start and end of thefts.
The footage presented challenges such as poor quality, non-sequential events from various stores, and sub-optimal camera angles, making accurate labelling and tracking difficult.
The Solution
Aya Data tackled these challenges by:
1. Pre-Processing the Footage: We divided the 5,000 hours of CCTV footage into manageable segments.
2. Experienced Annotation Team: Over 100 skilled video annotators labelled each frame, identifying individuals and marking theft events.
3. Quality Assurance: Each frame was manually annotated and reviewed by a quality assurance lead to ensure accuracy before being delivered to the client.
The Results
The meticulously labelled footage was used to train a custom computer vision model that successfully detects shoplifting and warns of suspicious behaviour. The client was fully satisfied with the accuracy and effectiveness of the solution.
Looking to improve your retail security with real-time theft detection?
Contact us today to discover how our advanced solutions can help safeguard your business and reduce losses.
Disclaimer: Aya Data respects client confidentiality and will not disclose any specific client details or project information. Any identifying information in our case studies may be anonymized.