Real-Time On-Site OpenGL-Based Object Speed Measuring Using Constant Sequential Image
Abstract
This thesis presents a method that can detect moving objects and measure their speed of movement, using a constant rate series of sequential images, such as video recordings. It uses the industry standard non-vendor specific OpenGL ES so can be implemented on any platform with OpenGL ES support. It can run on low-end embedded system as it uses simple and basic foundations based on a few assumptions to lowering the overall implementation complexity in OpenGL ES. It also does not require any special peripheral devices, so existing infrastructure can be used with minimal modification, which will further lower the cost of this system.
The sequential images are streamed from an IO device via the CPU into the GPU where a custom shader is used to detect changing pixels between frames to find potential moving objects. The GPU shader continues by measuring the pixel displacement of each object, and then maps this into a practical distance. These results are then sent back to the CPU for future processing.
The algorithm was tested on two real world traffic videos (720p video at 10 FPS) and it successfully extracted the speed data of road vehicles in view on a low-end embedded system (Raspberry Pi 4).