So before we delve into Khalil’s own test use of both YOLOv5 and Amazon SageMaker (using some of our past team event footage), let’s first look at object segmentation, detection and their uses.

Object detection and segmentation are two important tasks in computer vision that have a wide range of real-life applications. Object detection is the process that identifies the location of objects within an image or video stream, while segmentation involves dividing an image into multiple segments or regions based, on their visual properties.

One of the most common applications of object detection is in autonomous vehicles, where the technology is used to detect obstacles on the road, helping the vehicle navigate safely. Object detection can also be used in surveillance systems to detect and track individuals or objects of interest. In the retail industry, object detection can be used to track inventory and monitor shopper behaviour. Additionally, object detection and segmentation can be used in medical imaging to detect and diagnose diseases such as cancer.

As an Amazon Web Services (AWS) all-in consultancy, we work closely with a number of services related to object detection and segmentation. For example, Amazon Rekognition is a cloud-based image and video analysis service that can detect objects and scenes, as well as analyse faces and detect text. Amazon SageMaker is another AWS service that offers machine learning capabilities, including training your custom object detection and segmentation models that can be trained and deployed on AWS. That’s why Khalil opted for using SageMaker, as it works incredibly well with object detection and alternative software such as YOLOv5.

By leveraging these services, businesses can quickly and easily incorporate object detection and segmentation into their applications and workflows. For example, computer vision and object detection can be used in homes and commercial buildings for increased security and automated alerts when human motion detected. A healthcare provider could use object detection and segmentation to quickly and accurately analyse medical images, improving patient outcomes.

An example using our own test footage

To put this object detection work into practice, Khalil loaded up some footage from a previous Firemind team event. Footage that had a mixture of slowed movement as well as faster scenes, to present more of a challenge to YOLOv5. As you’ll see when you view the footage below, the model has been trained to recognise people and factor in a scoring based on recognisable facial features, body types, limb positioning and other data sets. These data types score a person as ‘1’ for a 100% match to trained data, to a ‘0’ (for no match – never occurs).

As you’ll see from the video, SageMaker and YOLOv5 does a remarkably good job at classifying the people within each scene, even when only partial elements of a person are visible. It also detects and segments each person within a reliable framework, for easy to view accuracy when viewing the footage.