JeVois  1.20
JeVois Smart Embedded Machine Vision Toolkit
Share this page:
Tips for writing machine vision modules

Handling exceptions

The JeVois framework fully embraces the use of exceptions. We believe that this provides for much more readable code, where the main flow is the one that is expected to work. This is in contrast to, for example, the Linux kernel, which is just a big pile of if statements to check whether some action succeeded or not.

In a Module process() function, the right thing to do if anything goes wrong is hence to throw and exception. The Engine will catch the exception and will cleanup and proceed to the next video frame, hopefully with better luck.

We recommend throwing std::runtime_error, as this will give more verbose messages to users (the error message will show the exception's what() string).

To be exception-safe and to avoid memory leaks, the easiest (and possibly best) is to always use exception-safe wrappers around dynamically-allocated memory. The most common is std::shared_ptr, which will automatically free the allocated memory when nobody is referencing it anymore.

Thus, instead of:

void process(jevois::InputFrame && inframe)
char * tmpdata = new char[100];
delete [] tmpdata;

which would leak the tmpdata memory in case of an exception, just use:

void process(jevois::InputFrame && inframe)
std::shared_ptr<char> tmpdata(new char[100]);

The shared_ptr will automatically free the memory when it runs out of scope (either because the process() function ends normally, or because of an exception). You can also just use an std::string or std::vector for tmpdata as those will free their memory when they run out of scope (including because an exception was thrown). std::unique_ptr would work here too.


JeVois may process images as various video resolutions. If it detects something in a frame and wants to communicate its coordinates to an Arduino, the problem arises of which resolution was being used and how to let the Arduino know.

For example, if an algorithm was processing video at 640x480, then an object detected at coordinates 320,240 is at the center of the field of view. But if the camera was grabbing at 320x240, now coordinates 320,240 are just past the bottom-right corner of the field of view. Just letting an Arduino know that something was detected at 320,240 could possibly be confusing.

Instead of having the Arduino do the work of supporting multiple camera resolutions, we provide in JeVois a standardized coordinate system:

See Helper functions to convert coordinates from camera resolution to standardized

Use imgToStd() to convert from image coordinates to the standardized coordinates.

Use stdToImg() for the converse.

Exception-safe wrapper around a raw camera input frame.
Definition: InputFrame.H:50