Problem

A Swedish-based client on Fiverr approached our team with a problem of shoplifting detection. They wanted to implement a solution that could detect shoplifters using state-of-the-art computer vision and machine learning techniques, to be able to sell it to stores. The client had access to some surveillance camera footage, but they lacked the expertise to develop an algorithm that could analyze the footage and identify potential shoplifters.

Our team recognized the challenge of developing an algorithm that could accurately detect shoplifting behavior in real-time. We knew that we needed to create a solution that would be both effective and efficient, while also being easy for the client to use.

Learn more about our approach here.

Solution

We began by analyzing the UCF crime dataset, which contained videos of various criminal activities, including shoplifting. Using this dataset, we trained a deep learning model using TensorFlow and Keras libraries. Our model was designed to identify specific behaviors associated with shoplifting, such as concealing items or attempting to leave without paying.

Once we had developed our model, we tested its performance and were pleased to see how efficient it is on real life shoplifting videos. With an AUC score of more than 75%, we exceeded the preset threshold of satisfaction of 70%.

Impact

Precision in detecting shoplifting events of 75,6%

"Ahmed did a great job and put a lot of effort in the work. We are very pleased with the resulting algorithm, which can now help us identify shoplifting and thus reduce incidents and flag suspicious activities."

Team

Team Member #1 - Machine Learning Engineer
- Analyzed the UCF crime dataset to identify behaviors associated with shoplifting
- Developed the deep learning model using TensorFlow and Keras libraries

Identify Shoplifting with AI Detection