Real-time Object Detection on Android using Tensorflow
Overview
Research shows that the detection of objects like a human eye has not been achieved with high accuracy using cameras and cameras cannot be replaced with a human eye. Detection refers to identification of an object or a person by training a model by itself. Detection of images or moving objects have been highly worked upon, and has been integrated and used in commercial, residential and industrial environments. But, most of the strategies and techniques have heavy limitations in the form of computational resources, lack of proper data analysis of the measured trained data, dependence of the motion of the objects, inability to differentiate one object from other, and also there is a concern over speed of the movement and Illuminacy. Hence, there is a need to draft, apply and recognize new techniques of detection that tackle the existing limitations.
Objective
A model based on Scalable Object Detection using Deep Neural Networks to localize and track people/cars/potted plants and many others in the camera preview in real-time. This is implemented in an android application and used handy in a mobile phone or any other smart device.
Motivation and State of Art
Humans learn to recognize objects or humans by learning starting from their birth. Same idea has been utilized by incorporating the intelligence by training into a camera using neural networks and TensorFlow. This enables to have the same intelligence in cameras, which can be used as an artificial eye and can be used in many areas such as surveillance, detection of objects/things etc.,
Literature Review
Currently, it is difficult to know when and where people occupy a building. The part of the difficulty arises due to the fact that the current sensor technology is not utilized to the full efficiency and also, due to the lack of utilization of proper data analysis methods. The comforting fact is that with the advent of 21st century, there has been a vast improvement in the sensor technology and arrival of the Internet of Thing (IoT) devices [1]. In an environment with clutter and noise, detection is even more challenging. One of the difficulties presented in literature for detection is to identify a human when stationary. This is a common problem for any sensor that is based on the reflection of acoustic, optical, or electromagnetic wave off a surface. However, the issue of stationary humans being identified as objects has been solved in my project. For example, fast moving objects in real time may cause confusion in identification or classification by computer vision techniques. At the tracking level, objects may be stationary for no apparent reasons or they may move in any direction spontaneously. This makes the tracking problem particularly challenging.
Furthermore, a particular type of technology may have difficulties meeting all necessary requirements in various lighting conditions, or rainy, foggy and inclement weather conditions, not to mention that most cameras or sensors have a limited field of view to monitor traffic in all directions. In addition, the clutter background and complex moving patterns of all objects on urban streets demand sophisticated and accurate real-time processing of sensor inputs to avoid false detection and recognition.
In order to overcome the aforementioned technical challenges, “You look only once” (Yolo) [2] detection system has been used not only to speed up the detection process, but also higher accuracy has been obtained. Yolo has not been implemented with android before and I have implemented this with android. One of the applications and advantages is that the android mobile devices are easily available with everyone, and in future, this detection system can be applied for Sousveillance [3].
Methodology
This is an enhanced version of https://github.com/tensorflow/tensorflow/tree/master/tensorflow/examples/android
