Research on Driving Intention Recognition Strategy Based on Accessible Area Detection and Drivers State Recognition in Off-Road Environments

Driving intention recognition is a frontier topic in intelligent transport. However less study involves the off-road scenario. Considering the complexity and random characteristic of off-road environment, the accessible space detection is a challenging item, it’s unreliable to only use facial recognition as inferred factor like on-road scenario. Therefore an anticipation strategy based on the fusion of freespace detection and driver facial recognition is proposed in this paper. This paper mainly consists of three parts, the first part involves the environment detection based on the Transformer network. In the Transformer framework, a cross attention mechanism is introduced to fuse the surface normal information and RGB image information. Meanwhile considering the particularity of off-road, we accomplish the dataset to train and test the network; The second part aims to design a face recognition algorithm based on the HOG features and cascade classifier. In order to solve the problem of camera out-of-focus caused by backlight, an optimization method is applied. In the third part we construct the driving intention recognition network based on the Fused Hidden Markov Model which can infer from the above two parts of feature sequences and recognize drivers’ intention of steering or changing lane. Proved by the recognition simulation result, our strategy has great accuracy and timeliness compared with other anticipation method such as SVM, LSTM, Random-Forest, etc.