Side Profile Vehicle Binary Images
Description
This is a dataset of vehicle's binary images obtained with an optical barrier. The main goal off this dataset is to train and test a vehicle classification method. Categories are defined according to the number of axles and the presence of double wheel or not. (See the vehicle_categories.png) This dataset is composed by 3 folders. Train: Images used to train a CNN. (Over 1000 images/cat) Total of 11232 images. Experiment1: Images used to validate.5389 images. Experiment2: Field test. Four days off images. 196018 images. Train and Experiment1 are selected images from the same lane, in different periods over 3 months. Experiment2 has images from several different lanes, and 3 different toll stations. We do not delete any image in this set.
Files
Steps to reproduce
We generate this images with an optical barrier installed in a Toll Lane. The optical barrier consists of two columns, approximately 1.80 m high, installed transversely to the vehicles' lane. A set of LEDs arranged at regular intervals emit parallel light beams from one column to the other. If there is no obstruction, the light beams are captured by a set of optical detectors installed in the other column on the opposite side. When a vehicle passes between the two columns that make up the optical barrier, the beams of light included in the region of the body and wheels will be blocked. However, the beams that are outside these regions reach the detectors. By sampling the light beams with sufficient frequency, a binary image of the vehicle side view profile can be constructed. We train an AlexNet CNN with the images in Train folder. We test the CNN with images in experiment1 and experiment2 folder. In the src folder there are three scripts made with matlab. The main script is the: 'train_and_test_deepLerning_alexnet_modified_fromScratch.m' It is used to train and test the network. You need to modify it and update the path to the images in your context It has some commented lines. By analyzing these lines it is possible to reproduce the different experiments.