Vehicle detection and statistics are an important component of computer vision. With the popular installation of traffic surveillance cameras, the use of computer vision for intelligent recognization and analysis of vehicles becomes very effective. Vehicle detection takes images as input, analyzes through a series of algorithms, and outputs labelled images with bounding boxes around the recognized vehicles.
Dataset (click here to access datasets)
The complexity of the dataset for this task is the video recorded on highways, vehicles with different relative distances, and variable traffic conditions. As for the size of datasets, the model training part using the training dataset which contains 1074 clips of 2s videos in 20 frames per second and 3222 annotated vehicles. Model evaluation part using the testing dataset which contains 269 clips of videos with the same format as training data.
The challenge of this technology is the complexity of the camera scene. When the vehicle appears in the photo at a far perspective, the target object occupies a small pixel, and the detection accuracy is low. When the vehicle appears in the close perspective in the photo, the target objects occupy large pixels; this causes a great change in the size of the target object.
I focus on the above issues to propose a viable solution, I apply the Histogram of Oriented Gradients to extract image features, select the support vector machine (SVM) as the classifier model to train the dataset.
- Preprocessing of Dataset
- Region of interest (ROI)
- Resize dataset
- Feature Extraction
- HOG feature extraction
- Classifier
- support vector machine (SVM)
- Object Detection
- sliding window
- Clone/Download the repo [https://github.com/conanzahn/Vehicle-Detection-Computer-Vision-.git]
- Download Datasets to your local machine [https://github.com/TuSimple/tusimple-benchmark]
- Run Jupyter notebook