Overview of Object Detection in Computer Vision


Reading time: 30 minutes

Object detection models are used to identify multiple relevant objects in a single image. The second significant advantage of object detection models versus image classification ones is that location of the objects is provided. Object localization algorithms not only label the class of an object, but also draw a bounding box around position of object in the image.Object Recognition uses CIFAR-10 dataset for Object Recogition.

Applications Of Object Detection Models

1.Most Important Application of Object Detection is self-driving cars to safely navigate through traffic

Object detection input image of road

The image essentially depicts that our car is near a square, and a handful of people are crossing the road in front of our car. As the traffic sign is not clearly visible, the car’s pedestrian detection system should identify exactly where the people are walking so that we can steer clear of them.

So what can the car’s system do to ensure this happens? What it can do is create a bounding box around these people, so that the system can pinpoint where in the image the people are, and then accordingly make a decision as to which path to take, in order to avoid any mishaps.

object detection road output

Our objective behind doing object detection is two folds:

1.To identify what all objects are present in the image and where they’re located

2.Filter out the object of attention spots violent behavior in a crowded place

3.assists sports teams analyze and build scouting reports

  1. ensures proper quality control of parts in manufacturing etc

Techniques to Split an Image and Use CNN

Approach 1

  1. Make a window of size much smaller than actual image size. Crop it and pass it to ConvNet (CNN) and have ConvNet make the predictions.
  2. Keep on sliding the window and pass the cropped images into ConvNet.
  3. After cropping all the portions of image with this window size, repeat all the steps again for a bit bigger window size. Again pass cropped images into ConvNet and let it make predictions.
  4. At the end, you will have a set of cropped regions which will have some object, together with class and bounding box of the object.But Passing so many Images To Convnet is computationaly more Expensive.the accuracy of bounding box is bad.

Approach 2

Another Algorithm which can be used in Object Detection is Divide and Conquer Algorithm.In this Algorithm we divide the image into parts and then feed them into image clssifier.It will Only predicts whether a particular Object is Present or not But we want a Bounding box over that object.

Approach 3

Best Approach of Object detection is using Deep Learning

1.Instead of taking patches from the original image, we can pass the original image through a neural network to reduce the dimensions

2.We could also use a neural network to suggest selective patches

  1. We can reinforce a deep learning algorithm to give predictions as close to the original bounding box as possible. This will ensure that the algorithm gives more tighter and finer bounding box predictions.This will Give the best Results.

Object Detection with sliding Window

In computer vision a sliding window is a rectangular region of fixed width and height that “slides” across an image.

object detection approach 1

Results

Object detection result

Popular Object Detection Models

  1. YOLO (You Only look Once)
  2. SSD (Single shot Detection)