Object Detection

Project Title

Object Detection

Project Area of Specialization Artificial IntelligenceProject Summary

Recently, deep ConvNets have signi?cantly improved image classi?cation and object detection accuracy. Compared to image classi?cation, object detection is a more challenging task that requires more complex methods to solve. There are many algorithms are exist to solve these problems but there have very few done on object detection for small objects, So our aim is to target this problem. Our project is mainly based on research and experimental analysis of existing algorithms.

Our project is to analyze and test some popular algorithms on datasets of small objects and observe their accuracies.

Project Objectives

We are targeting the drawbacks of previous existing algorithms such as R-CNN and Fast R-CNN, As these algorithms have overcome the storage and time issues of early models but still the improvements are very much possible in terms of detection of the small objects.

Our aim is to find the defects and calculate the accuracies of existing algorithms when they face small objects by this we could take effective steps in order to resolve these issues and developed improved algorithms.

Project Implementation Method

Fast R-CNN detection: 

Once a Fast R-CNN network is ?ne-tuned, detection amounts to little more than running a forward pass (assuming object proposals are pre-computed). The network takes as input an image (or an image pyramid, encoded as a list of images) and a list of R object proposals to score. At
test-time, R is typically around 2000, although we will consider cases in which it is larger (? 45k). When using an image pyramid, each RoI is assigned to the scale such that the scaled RoI is closest to 2242 pixels in area. For each test RoI r, the forward pass outputs a class posterior probability distribution p and a set of predicted bounding-box offsets relative to r (each of the K classes gets its own re?ned bounding-box prediction). We assign a detection con?dence to r for each object class k using the estimated probability Pr(class = k | r) ? = pk. We then perform non-maximum suppression independently for each class using the algorithm and settings from R-CNN. 

Benefits of the Project

Our project will beneficial for educational purpose as it is mainly based on research and experiments. But it is not limited, Our work would be used widely in future whenever anyone wants to work and go through the existing algorithms he will have to go through our work as our work will be targeting the week point of algorithms that is, detecting small objects. Our work reduced the important time and efforts of the researchers to indicate the defects of the algorithms which they will be studying and it will help others to develop advanced and efficient algorithms in the field of object detection which will be implied in many applications.

Technical Details of Final Deliverable

Our project will provide the results of existing algorithms but on new datasets, as they had not trained on that dataset before, we will train the models on visdrone dataset, which consists of drone view images. We will be training our model on multiple GPUs to reduce the training time, as without GPUs’s the deep learning models could take weeks in training because millions of images are used for training the model and we will train many models on visdrone data set and evaluate their results.

Final Deliverable of the Project Software SystemCore Industry SecurityOther Industries Education , IT , Others Core Technology Artificial Intelligence(AI)Other Technologies OthersSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	66000
GPU	Equipment	2	30000	60000
Thesis journals	Miscellaneous	5	1200	6000