Efficient deep learning framework for Automobiles rear end crash risk Prediction

Abstract

Safety consideration in transportation research plays a vital role newlinein Intelligent Transportation Systems. As Traffic accidents cause dangerous newlinehazards all around the world, a huge number of fatalities, serious social and newlineeconomic consequences and property losses have occurred. Collision Risk newlineIndex is a pre-accident risk assessment technique which is useful for newlinemonitoring crash risk. Perception Reaction Time (PRT) is one of the factors newlineto estimate the collision risk index in Rear End Crash Risk Index (RCRI). newlineIn the existing system, 1.5s is considered as average PRT. But, the newlinePerception Reaction time may be affected by individual driver s newlinecharacteristics, traffic characteristics, vehicle response time and newlineenvironmental conditions. To address this issue, Elastic Perception Reaction newlineTime (EPRT) is developed. Based on EPRT, Modified Crash Risk Index newline(MCRI) is constructed from RCRI to categorize the crash and non-crash newlineevents. newlineThe prediction and decision-making algorithms for collision newlineavoidance will be very beneficial to the drivers for making effective newlinedecisions in due time. The efficiency of the existing crash risk prediction newlinevi newlinealgorithms is diminished due to less accuracy. To overcome this issue, a newlinehybrid model Convolutional Neural Networks (CNN) and Long Short-Term newlineMemory (LSTM) with MCRI is presented for rear end crash risk prediction. newline98.1% of accuracy was obtained using CNN-LSTM with MCRI model. The newlinefindings reveal that MCRI has a high impact on the accuracy of the newlineproposed model. newlineFurthermore, the control parameters of the LSTM model have a newlinehigh impact on the efficiency of the Collision Risk Prediction (CRP) newlinesystem. Therefore, Improved LSTM (ImLSTM) was proposed where newlineLevenberg Marquardt is integrated with LSTM to optimize the newlinecontrol parameters of the LSTM model. CNN-ImLSTM model is used to newlinepredict the collision risk. Next Generation Simulation Project (NGSIM) is newlineused to evaluate CNN-ImLSTM model. Finding reveals that 98.4% newlineaccuracy was obtained for the proposed model. The CNN-ImLSTM model newlinehas a high accuracy