With the beginning of February, a new version of the NeuroCar red light violation detection system – NeuroCar RedLight 3 – was introduced to the NeuroCar product portfolio. The new version makes much wider use of artificial neural networks, including deep neural networks. A breakthrough is the use of a specialized neural coprocessor made by the Israeli Hailo-8, which significantly increases the efficiency of “on-edge” video processing and reduces energy consumption.
The NeuroCar RedLight system in version 3.0 is a continuation of a product of the same type, operating successfully on Polish roads since 2014. The main feature of the solution is the use of only the signal from multiple video cameras to detect, identify and document offenses. The use of on-edge processing and the fusion of data from multiple cameras allows for the detection of violations such as a vehicle running a red light, while maintaining high detection precision and high-quality documentation of the offense.
The new system uses various deep neural network models on a much larger scale. ANN models are used both to detect and track the license plate, as well as to recognize the registration number, country of origin, manufacturer and vehicle model version. In addition, deep networks are also used to track the vehicle in a view camera shot, in real time, detecting its movement trajectory. At the same time, all vehicles moving within the field of view of a measurement system consisting of a set of cameras observing vehicles at a single intersection inlet, both front and rear simultaneously, are tracked and identified.
The use of the Hailo-8 neural processor, mounted in a CPU located at the roadside, enabled a significant increase in computing power so that one CPU can now process the signal from 12 FullHD cameras simultaneously, at full speed of 25 frames per second.
The Hailo-8 CPU has negligible power requirements; it consumes no more than 4W of power at full load, which is more than 10 times better than GPUs with similar performance. The high power efficiency also allows for a significant reduction in heat generated by the CPU to is critical for on-edge applications.
