Computer Vision
Computer Vision, Spatial Detection, and SLAM are the fundamental layer that supports self-driving and mixed-reality applications
Computer vision based on deep learning model has been improved in all aspects of capabilities, efficiency and size, making it more accessible on edge devices and mobile devices.
Machine learning and computer vision has produced the magic in areas of: Image recognition and classification; Object detection and tracking; Semantic and instance segmentation; Image enhancement and generative art; Reinforcement learning to understand surroundings, etc.
DeepRacer: get it ready to learn to drive
Makeable club members created EverySecond system to help elderly people.
The wearable camera device, as a watchful eye for the elderly. The object detection and event construction algorithm, as the brain to understand and remember what has happened.
Deploy tinyML CV model on Raspberry Pi 4, for image classification and object recognition
Object detection model on Jetson Nano, to count and track workshop attendance.
ML based text recognition on mobile device, extracting credit card number and to retrieve corresponding account information
Object tracking and image transformation on mobile device
DeepRacer: learning to drive on my backyard basketball court

Build

A note about EverySecond:

A couple of high school students at Makeable Club wanted to make something to help their grandparents with daily lives and safety, and they came up with this EverySecond idea - a digital short term memory for the elderly, so they can recall what has happened or not during the day, such as "have I taken my medicine pills?"

I mentored the students and pointed the direction to Google Cloud for them to learn the AutoML for training their own object detection model, and learn the serverless computing for image recoginition and memory recall functions.

They created a wearable camera device to record daily activities from First-Person-View, used their custom-trained object detection model and inference to understand what have happened and saved them in a Data Store as short-memory, then set up a website as an interface to recall the "short-term memory" by easy queries.

One of the EverySecond creators, Emily, talks about the making of EverySecond at a technology company in the Bay Area.