Defense Unicorns - Embedded Systems Practical

Overview

This template repository contains a bare skeleton of a C++ project meant to evaluate an Embedded Systems Programmer's hands-on abilities. The exercise is specified below in "Background and Exercise"

Please spend no more than 4 hours on the exercise. The goal is a working application that is free from defects such as memory allocation errors, etc.

To complete the exercise please create a new repository from this template and push your changes to it. When complete email your recruiting contact at Defense Unicorns. The hiring team will then evaluate your submission, and if it meets the standard, move onto an in-person review of the submission where enhancements and other changes can be discussed.

Background and Exercise

Image data acquisition is desired, but no software interfaces for it exist. Hence, we need to write a program to read images from a raw stream.

The raw stream consists of:

• A uint12 (12 bit) value per pixel. They represent the grayscale reading for a pixel. A value of 0 means no signal, whereas a value of 4095 means saturation.
• There are 640x480 = 307,200 pixels represented per reading. It's a rolling shutter camera scanning horizontally. Hence, the first 640 values in the stream represent the top most row in the image. The second 640 values represent the second from the top row in an image
• There is no padding between values. In other words, one capture of an image consists of 640*480*12/8 = 460,800 bytes.

Attached is a raw stream dump of the sensor readings. Please feel free to compile as you go.

1. Load this data into memory (filename is image_binary_dump).

2. Create a function to access pixel values by x,y coordinates. For example, (x=10,y=200) corresponds to the 10th pixel from the left, and 200th pixel from the top. The bit significance of each uint12 pixel is 11,10,9,8,7,6,5,4,3,2,1,0.

3. A small crop of the image needs to be passed through. Please write a function that can crop out a box from this data. It takes 4 parameters: x,y,w,h. The x,y is the upper left starting location of the subset, w is width of the box (in pixels), and h is the height of the box (in pixels). Crop the image using the following parameters (x=208, y=110, w=418, h=296).

4. Often times, ML models consume a floating-point representation. The float value of each pixel is linearly dependent on the uint16 value, where a fully saturated pixel maps to a value of 1.0 and a non-activated pixel maps to a value of 0.0. Convert the image to a float image. Don't worry about which type of float.

5. Output the cropped float image to stdout.

6. Given this will be a video stream, we need to constantly load (#1), crop (#3), convert (#4), output (#5), unload, and repeat for the next frame - design functionality for this. It's assumed the resolution of a single image will not change. (filename is video_binary_dump).

7. How would you validate the results and the code you wrote?

Tooling

Included in the exercise is a VSCode-compatible development container configuration that includes a working C++ environment. You are encouraged but NOT required to use the devcontainer in the course of your work as the *engineers assessing your work will expect build and run to work in the *environment.

The included Makefile's default target builds and runs the included binary out of the box.

Submission

If needed, please update the included Makefile and build include any custom build instructions. Acceptable submissions must be built via make build and make run and no more.

Please compile and git commit as you go - your git history is also part of your submission.

Any compiler or operating system is OK, but GNU/Linux is preferred. Please use only the stdlib for questions 1 to 6. You can use whatever libraries (if wanted) for #7.

The final deliverable is a tarball (optionally compressed) containing this directory with any changes but excluding any binary artifacts (object files or executables).