TraffiKAI

Nishank K S

Bengaluru, Karnataka

0 0

A-EYE on ROADS! An AI & ML solution to solve one of the basic but most important traffic problems in day to day life. ...learn more

Project status: Published/In Market

oneAPI, Artificial Intelligence

Intel Technologies
DevCloud, oneAPI, Intel Opt ML/DL Framework, Intel Python, OpenVINO, Intel CPU

Docs/PDFs [1]Code Samples [1]Links [5]

Overview / Usage

Problem Statement: The increasing number of vehicles in cities can cause high volume of traffic, and implies that traffic congestion has become more critical nowadays. Fatalities due to traffic delays of emergency vehicles such as ambulance & fire brigade is a huge problem. In daily life, we often see that emergency vehicles face difficulty in passing through traffic.

What data says?

According to Times of India about 146,133 people were killed in road accidents in India in the year 2016. Unfortunately about 30% of deaths are caused due to delayed ambulance.
According to the Radhee Disaster and Education Foundation, one in 10 patients in India dies on the way to hospital.
Another Indian government data shows that more than 50% of heart attack cases reach hospital late , which can constitute unavailability of ambulances too but majority of it is due to patients stuck in traffic.
The statistics show that, in cities with the worst traffic in the world Mumbai is ranked 5th, Bangalore is ranked 10th and Delhi is ranked 11th.

Methodology / Approach

Objective: Objective of proposed solution is to improve efficiency of existing traffic signaling system. The goal of the project is to automate the traffic signal system and make it easy for the traffic police department to monitor the traffic.

Solution: The solution to solve the above problems as proposed are Dynamic Traffic Signaling and Emergency Vehicle Detection through both audio and video. The aim is to keep the same infrastructure and make delta changes in the system using the power of AI & ML.

Dynamic Traffic Signaling

Dynamic Traffic Signaling is implemented by calculating the density of traffic in each lane in a multi lane system and using this information it turns the signal lights green or red accordingly. It allocates the least time to the lane which has less density traffic and the time saved here is allocated to the lane which has high density traffic.
Object detection algorithm: Single Shot Detector (trained on COCO dataset)
Tech Stack: Python, PyQT, OpenCV, Streamlit

Emergency Vehicle Detection

Emergency Vehicle is detected by two methods in order to ensure the certainty of presence of an emergency vehicle in the input medium. The two methods include audio and video. Firstly, the video is processed frame by frame and the presence of emergency vehicles are found out and returns the confidence level which inturn returns a probability score. The detection is also preformed through audio and the video's audio is passed through a CNN model which gives a probability score. The probability scores from each models is obtained and ensemble learning is performed to get the final verdict.
Image Classification algorithm: DenseNet-169 Teck Stack: Python, PyQT, OpenCV, Streamlit

Technologies Used

TECH STACK

Audio detection model: Custom model using CNN with LSTM layer
Video detection model: DenseNet-169, ResNet 50
Object detection model: Single Shot Detector (Trained on the COCO Dataset)
GUI: PyQt5, Streamlit
Libraries: Tensorflow, OpenCV, PyTorch

Toolkit used: Intel® AI Analytics Toolkit (AI Kit) - oneDNN (Deep Neural Network Library)

TraffiKAI uses multiple memory intensive machine learning models which increases the runtime by a significant amount causing a delay in the processing of the input videos on the systems with limited processing power.
The Intel® AI Analytics Toolkit (AI Kit) helps in achieving better results by optimising the models with the help of oneAPI Deep Neural Network Library (oneDNN).
TraffiKAI uses state-of-the-art deep learning frameworks like PyTorch and Tensorflow which are optimized for the Intel architecture by the oneAPI platform and further boosts the inference of the models.
scikit-learn is an important library which provides various algorithms of machine learning as functions. Intel(R) Extension for Scikit-Learn is also enabled to improve the performance.
The toolkit also has support for a number of pre-trained models such as DenseNet-169, YOLOv3, LSTM (audio) which are used in TraffiKAI and help to improve the performance. Using the pre-trained models, transfer learning has been implemented on the Intel DevCloud for oneAPI which improvises the performance and accuracy.
Intel® Distribution of OpenVINO™ Toolkit is also used to boost the object detection models.