Suraj Pattar

Robotics AI Research Engineer / Ph.D. in AI and Robotics / Data Scientist / 3D Printing Enthusiast

Vehicle Detection Comparison

Keywords: Pedestrian Detection, Computer Vision, GTA5, OpenCV, MATLAB, Yolov4, Detectron2

Project Description

Comparison video for Vehicle Detection in GTA5 with Matlab(Vehicle Detector ACF), OpenCV (Haar Cascades), Yolov4 and Detectron2 (MaskRCNN). I was trying to compare how well the models generalize to unseen synthetic data.

Matlab Vehicle Detector

Matlab’s Vehicle Detector ACF is a detector using Aggregate Channel Features. I make use of the ‘full-view’ model which is trained using unoccluded images of the front, rear, left, and right sides of the vehicles.

OpenCV Cascade Classifier

OpenCV Haar Cascades classifier is a machine learning based approach where a cascade function is trained from a lot of positive and negative images. It is then used to detect objects in other images. Also known as the Viola-Jones algorithm. I make use of a pretrained ‘cars.xml’ file which is trained on images of cars.


YoloV4 is an improved version of Yolov3, with the implementation of a new architecture in the “backbone” and the “neck”, which has improved its mAP (Mean Average Precision) on the COCO dataset by 10% and the inference speed in FPS by 12%.


Detectron2 is Facebook AI Research’s next generation software system that implements state-of-the-art object detection algorithms. It is a ground-up rewrite of the previous version, Detectron, and it originates from maskrcnn-benchmark. I make use of the MaskRCNN model.

(Link to the code coming soon).