A Strategic Framework for Reducing Decision Bias in Driverless Car Object Detection

Bias in machine learning is a significant problem that demands industry-wide attention, and even more so in the case of computer vision applications. The debate is whether autonomous vehicles are safe, yet more likely to strike a pedestrian of color than a person of white skin color. It is essential to obtain a greater understanding of the algorithmic bias that occurs during driving-centric object recognition. This research attempts to understand algorithmic bias that emerges during driving-centric object recognition. This project is opportune since several jurisdictions have recently permitted autonomous cars to run on public roads and several businesses are testing their AVs on public roads, often resulting in a notable number of fatalities. Autonomous systems are vulnerable to algorithmic bias because they do not need an active or passive person to identify and correct such biases. To reduce bias, design and implementation methodologies can be enhanced in accordance with new ethical norms. The aim is to continue efforts to reduce object detection bias that affects autonomous system decision-making to promote safety and increase technology adoption. Autonomous cars must evaluate safe operating options when a collision can't be prevented. This may persist due to the fluidity and unpredictability of human-automobile interactions. Artificial intelligence (AI) systems are biased because to their reliance on data, individuals, and model design. Data bias has been connected to obstacles associated with the deployment of ethical AI and has detrimental effects on underrepresented demographic groups. This study addresses training data bias, algorithmic processing bias, transfer context bias, and interpretation bias. The first two biases relate to the computational aspects of the problem. The latter two relate to the improper deployment of algorithms and autonomous systems. If the training data are condition-specific, the model cannot make precise judgments in the actual world. This research studies and reviews strategies that would mitigate object detection biases in an autonomous vehicle system, specifically whether several simple building blocks used for research-grade pedestrian detection have comparable ability to detect pedestrians with cognitive and physical disabilities.