Brand Detection and Video Recognition
Artificial Intelligence (AI) can be superior to people wherever patterns in text, video, image or audio files need to be identified correctly. Thus, the question of how to use Artificial Intelligence to recognize images has become very popular and is now a well-understood problem. Even cloud and AI leaders have developed their individual cognitive service APIs for that purpose, e.g., Amazon (Amazon Recognition), Microsoft (Microsoft Computer Vision), Google (Google Cloud Vision) and SAP (SAP Leonardo Machine Learning). We at Arvato Systems worked on not only solving the technical challenge but also on connecting AI to real life business benefits. As introduced in our June newsletter we used AI and video recognition to revolutionize the way returns are received and managed. Out of that first proof of concept (POC), we have recognized the need to understand how AI can analyze images taken on smartphones and detect the displayed brand. This question became an even more significant challenge as we leveraged AI to not only recognized brands on images but also in presented videos.
Brand recognition in videos with the use of AI
The showcase brand recognition in videos with the use of AI demonstrates the ability of Machine Learning to analyze video footage in terms of product identification, product localization, and product tracking. Therefore, AI and machine learning are excellently suited for companies to improve their processes in areas such as marketing or product returns. One marketing focused use case could be the use of AI to analyze product placement in movies or sports advertisements. For that, a neural network is previously trained on brand logos (e.g., Nike, Audi, or other objects or products) and viewed video material is searched for those brands. At the same time, the neural network will recognize when and where a brand was seen in the image and how long that brand was visible.
With this new technology, companies could gain more detailed knowledge around their efforts and the received data can be used to improve the commercial or overall branding strategy.
Benefits of brand recognition include:
Our new brand and video recognition POC showed the possibilities to connect custom models with provided APIs models to generate a higher business effect. The development of the image recognition services by Amazon, Microsoft, SAP, and Google show that AI technology is seen as significant benefit and gamechanger when it comes to the automation of image recognition in real business cases such as return centers. As remarked by google "the question is not if this technology will be part of a product. The question is which product will not use AI.”
The Benefits of This POC in Summary:
More details and technical information
To solve the challenge and achieve our goal also to detect brands in videos, we have leveraged in-house custom-developed models which can be combined with a diverse set of third-party cognitive service video APIs such as AWS SageMaker, Microsoft Azure Custom Vision Service. Unfortunately, Google’s services could not be used for the further development of the original POC as Google Cloud AutoML does not feature a trainable object detection algorithm yet.
We have managed that a fixed set of brands and clothing pieces is detected in images taken by a webcam or the camera of a smartphone by using two separate machine learning services (Yolov3 and Azure Custom Vision Service).
To also leverage AI to recognize brands in videos, we have implemented the application with Angular. Videos can be uploaded, and images are then extracted from the video. The pictures are then sent to the cloud APIs and resulting bounding boxes of the detected brands are overlaid onto a video player. Those overlaid boxes are then analyzing not only the brand but also the total time a brand was visible during the video is displayed.
For further questions please feel free to contact Meikel Bode.
