Haar-Like feature implementation on FPGAs for Advanced Driver-Assistant Systems

Nowadays, there is a growing trend in developing advanced computational techniques for Advanced DriverAssistant Systems (ADAS) using low-cost devices. One of the relevant ADAS application is vehicle detection based on camera sensors. The state-of-the-art approaches for this application utilize machine learning techniques. These techniques might be divided into feature extraction and inference algorithms. Feature extraction requires high computational capabilities and execution time because it extracts the most relevant data from a camera frame. From the embedded system point of view, lowcost heterogeneous systems based on FPGAs might provide a framework to develop feature extraction techniques. According to vehicle detection, haar-like is one of the most used feature extraction technique in the literature. As a result, this paper proposes a haar-like feature extraction architecture in hardware for achieving real-time requirements in the automobile industry, analyzing parameters as resource utilization and system latency.