Leveraging Video Analytics – An Insight to Emotion Analysis


As mentioned in my previous blog “Video Analytics: The What, Where, Why and How”, video analytics can be a game changer in the real-life scenarios. Read the first part of this series here: 

It’s not what you say, but how you say it !

Facial expressions are really fundamental to interpersonal communication and can convey a great deal about the state of mind. Human emotions can be broadly categorized as – happiness, sadness, fear, disgust, anger, surprise and neutral. Industries leverage human emotions to understand customers’ experience and their journey through the lifecycle of products and services that are being offered.

Emotion Analytics is a state-of-the-art technology that identifies and analyzes the band of the human emotional spectrum including moods, attitudes, the way of talking and personality. Coupled with the explosive amount of data available, businesses are eager to utilize the exemplary growth of Artificial Intelligence. There are umpteen use cases for businesses using this scope, a few among them are emotion detection, emotion analytics, people counter and demographic analysis etc.

The market research says –  “The global Emotion Analytics Market is expected to grow at USD ~25 billion by 2023 at a CAGR of ~17% during the forecast period 2017-2023.Global leaders such as Apple, Microsoft, Retinad Virtual Reality and Neuromore have their own Emotion Analytics software and many are joining the wagon. It’s time for more companies to invest in this space to improve their marketing, sales, services, and customer experience and unleash the true potential of data. 

Video content analysis serves well for businesses where customer interactions are high and customer satisfaction plays a major role in expanding the business.

For e.g. Let’s consider the cameras placed in the retail stores, the camera captures the emotions of the people while they shop, this helps in answering some critical questions like-

  • Does the customer smile while seeing the product?
  • Is a curious expression given while looking at the product?
  • Does the product disappoint them?
  • How much time do they spend on a particular aisle?
  • Are they relaxed or anxious at the billing counter?

Using all this information in  the live video feeds for tracking people. An ID would be assigned for all possible detections in a frame. In the subsequent frames, the person’s ID is carried forward. If the person has moved away from the frame then that ID will be dropped. Also if  a new person appears then the process starts off with a fresh ID.

At Latent View Analytics, we have provided a solution for a large restaurant chain by implementing the people counter and emotion detection to improve their customer experience metrics. This application highlights the number of people in the queue outside the restaurant and also helps in detecting their emotions.

The application uses the YOLOv3 as a neural network which is trained on darknet architecture. It can localize multiple objects and also predicts the classes that the objects belong to (house, horse, ball, car, apple etc.).

 A single neural network is applied to the full image. This network divides the image into regions and predicts bounding boxes and probabilities for each region. These bounding boxes are weighted by the predicted probabilities.

In the below image, the bounding boxes and the probabilities for the class ‘Person’ are seen.

Crowd/People counting mechanism example:

41.2 1
41.2 1

Emotion Detection example: In the images below, the bounding boxes and the prediction for multiple classes of emotions are shown.

Video Analytics can add a lot of value to the business by optimally utilizing the data being generated by business and by improving customer loyalty. Some of the many value adds of video content analysis are –

  1. Identifying Hot and Cold Zones – the analytics model can divide the store space into hot and cold zones based on foot traffic and amount of time spent in these zones. This will help them optimize their inventory positioning, marketing campaigns, store layout maps etc.
  2. Queue Management – wait times at point of sale can be reduced by a better pattern recognition.
  3. Gauging Customer Behaviour – Analytics will help determine customers’ product exposure level, engagement, frequency, area of interest etc.
  4. Customer Service – Analytics will help in knowing the busiest periods, balance staff resources and better store management 
  5. Retail Shrinkage/Loss Prevention – The model is trained to detect unusual activities like unexpected time of operation, unauthorized access, suspicious movement of inventory and more. 

Video Analytics, when adopted by businesses around the world will help them add a new dimension to their existing analytics efforts.

Related Blogs