Identification of Driver Braking Intention Based on Long Short-Term Memory (LSTM) Network

Abstract

Driving intention identification is a key technology which can improve the adaptability of the intelligent driver assistance systems and the energy efficiency of electric vehicles. This article proposes a novel method for identifying the driver braking intention. In order to improve the identification accuracy of driving intention, a braking intention identification model based on Long Short-Term Memory (LSTM) Network is constructed. The data of slight braking, normal braking and hard braking that can use for offline training are obtained through tests on real vehicle at Chang'an University vehicle performance testing ground. Support vector machine-recursive feature elimination (SVM-RFE) algorithm is used to select the characteristic parameter of braking intention identification model. The random search is subsequently used to optimize the hyper-parameters of LSTM. LSTM-based and General Hidden Markov Model (GHMM)-based model under different time window are used to identify braking intention of slight braking, normal braking and hard braking respectively. The results show that the Precision, Recall, F-measure, Accuracy of the braking intention identification model which propose in this paper based on LSTM are better than that of the braking intention identification model based on GHMM. Moreover, the Recall and Accuracy of the LSTM-based braking intention identification models are above 0.95, indicating the good ability of intention identification.

Selection of Characteristics by Hybrid Method: RFE, Ridge, Lasso, and Bayesian for the Power Forecast for a Photovoltaic System

“…Comparing the results with those obtained by [21] who proposes another hybrid model formed by a combination of RFE and SVM to identify braking intentions in electric vehicles based on a short-term memory network (LSTM) and Gaussian Hidden Markov Model (GHMM), obtaining a precision of 95%, In the same way for the estimation of the contractile function of the left ventricle, a value necessary to carry out a successful heart transplant, [22] also uses a combination of SVR and RFE to select the estimation parameters, reaches a value of 88%. In the same way [23] proposes another hybrid model that uses principal component analysis (PCA) and RFE to select the frequency spectral coefficients and then classify them with a multilayer perceptron network for the determination of age through the characteristics of speech.…”

Review and Perspectives on Driver Digital Twin and Its Enabling Technologies for Intelligent Vehicles

“…In addition to the external driver state information, the vehicle needs to understand the driver's intention to generate appropriate assistance and collaborative control strategies. Current studies focus on specific tasks or scenarios such as lane change intention [136][137][138], braking intention [139][140][141], and acceleration intention [142], and they can be summarized as shown in Fig. 6.…”

“…[142] proposed an intention-oriented model for longitudinal dynamics based on the commonly available signals on the CAN bus of modern vehicles. [139] presented a novel braking intention identification model based on the LSTM network to recognize three levels of braking: slight, normal, and hard, in which braking-related data obtained from the speed sensor, gyroscope, and pedal force sensor, were utilized to achieve a predictive accuracy greater than 95%. Further, the driver's cognitive signal was investigated to detect the intention to perform emergency braking [140,141], with an experimental accuracy greater than 90%.…”

A Study on a HMM-Based State Machine Approach for Lane Changing Behavior Recognition

“…Features used to describe a particular driving situation play an important role in driving behavior predictions. Thus, methods such as filter and wrapper methods are used in [9] to select the most appropriate features as input variables in the prediction and recognition models. The wrapper method considered in [9] employs the combination of SVM with a recursive feature elimination method to extract features [10], [11].…”