Objective: The safety enhancement of road users has begun to gain more attention, in particular the innovation and application of ADAS. The accurate and timely detection of the risk of accident has become an active area of research, with the focus on the drivers and other vulnerable road users.
The neuromorphic visual information processing method, inspired by Hubel and Wiesel’s experiments on mammalian visual cortex, is proposed as a possible solution to these tasks. The proposed method replicates the performance of visual cortex in practical computing settings. By applying the orientation feature extraction and subsequently applying the neural network ensured robustness and accuracy.
Method: The proposed system has been evaluated on pedestrians/cyclists detection and driver monitoring, with a particular focus on emotion/stress detection. The tests have been carried out with video data sets of various conditions, with the experimentation and data set generation at public roads in every day settings.
The neuromorphic visual monitoring of drivers for the attentive or emotional status has been also evaluated, as approximately 15% of road accidents have been caused by the dangerous driving in ‘anger and or/frustration’. The driver monitoring system by detecting the emotional state from the limited facial image of driver would make the measures of early warning against possible dangerous or inattentive driving. The neuromorphic system was evaluated to determine the warning signal based on the emotional state detection, based on the key feature extracted from the face images. The test was based on the facial database (JAFFE) of six basic emotional states.
Results and Conclusion: The performances of neuromorphic visual information system were measured to the success rate 99% of pedestrian/cyclist detection, and the successful recognition 91% of facial emotional states. The real-time performance was evaluated with the neuromorphic ASIC, fabricated by the automotive CMOS technology. The processing speed of neuromorphic ASIC alone was tested for the speed of 30 frames per second, without the latency or external memory.